CINXE.COM

<!DOCTYPE html> <html lang="ja" data-page-scope="EntrySearch" data-stable-request-url="https://b.hatena.ne.jp/q/SHAP" data-device-type="PC" data-sentry-environment="production" data-sentry-sample-rate="0.1" data-entry-search-query="SHAP" data-page-subtype="entrysearch" data-page-type="entrysearch" > <head>  <script> (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start': new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0], j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src= 'https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f); })(window,document,'script','dataLayer','GTM-5TDHQX'); </script>  <meta charset="UTF-8"> <title>SHAPの人気記事 29件 - はてなブックマーク</title> <script src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/js/v4/bookmark.js" async></script> <link type="text/css" rel="stylesheet" href="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/css/v4/pikaday.css" /> <link type="text/css" rel="stylesheet" href="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/css/v4/triangle.css" /> <link type="text/css" rel="stylesheet" href="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/css/v4/bookmark.css" /> <link type="text/css" rel="stylesheet" href="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/css/v4/bookmark.star.css" /> <link rel="canonical" href="https://b.hatena.ne.jp/q/SHAP" /> <link rel="shortcut icon" type="image/x-icon" href="/favicon.ico" /> <meta name="viewport" content="width=device-width"> <link rel="search" type="application/opensearchdescription+xml" title="はてなブックマーク検索" href="/opensearch.xml" /> <link rel="apple-touch-icon-precomposed" href="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/apple-touch-icon-precomposed.png" /> <link rel="mask-icon" href="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/safari-pinned-tab-icon.svg" color="#00A4DE" /> <meta name="msapplication-navbutton-color" content="#2C6EBD" /> <meta name="msapplication-task" content="name=はてなブックマーク; action-uri=/; icon-uri=https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/icons/bookmark.ico" /> <meta name="msapplication-task" content="name=マイブックマーク; action-uri=/my; icon-uri=https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/icons/my-bookmark.ico" /> <meta name="msapplication-task" content="name=お気に入りのブックマーク; action-uri=/my/favorite; icon-uri=https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/icons/favorite-bookmark.ico" /> <meta name="msapplication-task" content="name=人気エントリー; action-uri=/hotentry; icon-uri=https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/icons/hotentry.ico" /> <meta http-equiv="X-UA-Compatible" content="IE=edge"> <meta name="referrer" content="origin"> <link rel="alternate" type="application/rss+xml" href="/q/SHAP?sort=popular&safe=on&mode=rss&users=3&date_range=5y&target=title" /> <script> function isMobile() { const html = document.documentElement; const deviceType = html.dataset.deviceType || ''; if (deviceType.length > 0) { return deviceType === 'MOBILE'; } else { const maxMobileWidth = 480; const windowWidth = window.innerWidth; return maxMobileWidth >= windowWidth; } } </script> </head> <body class="page-centerarticle page-entrysearch fixed-header">  <noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-5TDHQX" height="0" width="0" style="display:none;visibility:hidden"></iframe></noscript>  <div class="link-app-banner js-dynamic-link-app-banner is-hidden"> <button class="link-app-banner-btn-close js-dynamic-link-app-banner-close-button"><span>close</span></button> <a class="link-app-banner-btn" href="https://hatenabookmark.page.link/app"> <div class="link-app-banner-btn-detail"> <div class="link-app-banner-icon"><img src="/images/v4/public/app/ic-ios.svg" alt="はてなブックマークアプリ"></div> <p class="link-app-banner-text">サクサク読めて、<br class="link-app-banner-text-newline">アプリ限定の機能も多数！</p> </div> <div class="link-app-banner-btn-open">アプリで開く</div> </a> </div> <header id="global-header" data-portal-static-origin="//cdn.www.st-hatena.com"> <div id="header" class="global-header-inner js-global-header-inner"> <h2 id="gh-logo"><a href="/" data-gtm-label="gh-blogo"><span>はてなブックマーク</span></a></h2> <div class="is-guest js-guest is-hidden"> <input id="gh-search-icon-check" type="checkbox" checked="checked" /> <label class="gh-search-icon" for="gh-search-icon-check" aria-label="検索を閉じる"><span>閉じる</span></label> <form class="gh-searchbox" method="get" action="/search_dwim" role="search"> <input value="" name="q" type="search" class="gh-inputtext" placeholder="キーワード・URLを検索" size="40" /> <input value="" type="submit" class="gh-search-button" /> </form> <ul class="gh-service-menu"> <li><a href="/guide" data-gtm-label="gh-guest-guide">はてなブックマークって？</a></li> <li><a href="/guide/tools" data-gtm-label="gh-guest-tools">アプリ・拡張の紹介</a></li> <li><a href="https://www.hatena.ne.jp/register?location=https%3A%2F%2Fb.hatena.ne.jp%2F%3Fregistered_from%3Dglobal-header&via=201037" class="gh-guest-register" data-gtm-label="gh-guest-register">ユーザー登録</a></li> <li><a href="https://www.hatena.ne.jp/login?location=https%3A%2F%2Fb.hatena.ne.jp%2Fq%2FSHAP" rel="nofollow" class="gh-guest-login" data-gtm-label="gh-guest-login">ログイン</a></li> <li class="gh-hatena-logo"><a href="http://www.hatena.ne.jp/"><img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/logo/logo-global-white.svg" alt="Hatena" width="74px" height="13px"/></a></li> </ul> </div> <noscript> <ul class="gh-service-menu"> <li><a href="https://www.hatena.ne.jp/logout?location=https%3A%2F%2Fb.hatena.ne.jp%2F">ログアウト</a></li> </ul> </noscript> <script type="text/x-template" id="template-menu-hatena-user"> <form class="gh-searchbox" method="get" action="/search_dwim" role="search"> <input value="" name="q" type="search" class="gh-inputtext" placeholder="キーワード・URLを検索" size="40" /> <input value="" type="submit" class="gh-search-button" /> </form> <ul class="gh-service-menu"> <li><a class="hotentry" data-gtm-label="gh-mypage" href="{{hotentry_url}}"><img src="{{user_image_url}}" class="header-profile-icon" width="16" height="16" alt="{{username}}" /></a><a class="hotentry" data-gtm-label="gh-mypage" href="{{hotentry_url}}">マイページ</a></li> <li><a class="bookmark" data-gtm-label="gh-bookmark" href="{{bookmark_url}}">ブックマーク</a></li> <li><a class="unread_bookmark" data-gtm-label="gh-ril" href="{{bookmark_stock_url}}">あとで読む</a></li> <li><a class="add" data-gtm-label="gh-add" href="{{add_bookmark_url}}">追加</a></li> <li><a class="tools" data-gtm-label="gh-tools" href="/guide/tools">ツール</a></li> <li><a class="config" data-gtm-label="gh-config" href="/-/my/config/profile">設定</a></li> <li><a class="feedback" data-gtm-label="gh-feedback" href="/-/feedback/hatena_bookmark" target="_blank">フィードバック</a></li> <li id="header-username" class="gh-dropdown-services" data-gtm-label="gh-services" title="利用中のサービス" data-name="{{username}}"> <span class="header-dropdown gh-dropdown" tabindex="0"> <img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/common/usermenu-wh.svg" alt="利用中のサービス" width="24" height="24" /> </span> <dl class="header-window" id="username-window"></dl> </li> <li id="header-notify" data-gtm-label="gh-notify" class="gh-dropdown-notify" title="あなたへのお知らせ"> <span class="header-dropdown gh-dropdown" tabindex="0"> <img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/common/notify-wh.svg" alt="あなたへのお知らせ" width="25" height="25" /> <span class="notify-count"></span> </span> <div id="notify-window" class="header-window header-notify"></div> </li> <li class="gh-hatena-logo"><a href="http://www.hatena.ne.jp/"><img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/logo/logo-global-white.svg" alt="Hatena" width="74px" height="13px"/></a></li> </ul> </script> </div> </header> <div id="info-header"><div id="info-header-message"></div><span></span></div> <div class="global-notification is-hidden js-global-notification"></div> <div id="container"> <div class="entrysearch-searchbox"> <form class="entrysearch-searchbox-body js-entrysearch-form" method="get"> <input value="SHAP" type="text" class="entrysearch-searchbox-textInput js-entrysearch-form-text-input" size="40" aria-label="検索キーワードを入力"> <input value="title" name="target" type="hidden"> <input value="popular" name="sort" type="hidden"> <input value="3" name="users" type="hidden"> <input value="on" name="safe" type="hidden"> <input value="検索" type="submit" class="entrysearch-searchbox-btn js-entrysearch-form-button" aria-label="検索"> </form> </div> <div class="centerarticle-wrapper"> <div class="left-container"> <div class="centerarticle-sidebar-menu"> <h3 class="centerarticle-aside-title">検索対象</h3> <ul class="centerarticle-sub-navi"> <li><a href="/q/SHAP?safe=on&sort=popular&target=all&date_range=5y&users=3" >すべて</a></li> <li><a href="/q/SHAP?target=tag&users=3&date_range=5y&safe=on&sort=popular" >タグ</a></li> <li><a href="/q/SHAP?safe=on&sort=popular&target=title&date_range=5y&users=3" >タイトル</a></li> <li><a href="/q/SHAP?safe=on&sort=popular&target=text&users=3&date_range=5y" >本文</a></li> </ul> </div> <div class="centerarticle-sidebar-menu"> <h3 class="centerarticle-aside-title">並び順</h3> <ul class="centerarticle-sub-navi"> <li><a href="/q/SHAP?target=title&users=3&date_range=5y&safe=on&sort=recent" >新着</a></li> <li><a href="/q/SHAP?users=3&date_range=5y&target=title&sort=popular&safe=on" class="is-current">人気</a></li> </ul> </div> <div class="centerarticle-sidebar-menu"> <h3 class="centerarticle-aside-title">ブックマーク数</h3> <ul class="centerarticle-sub-navi"> <li><a href="/q/SHAP?safe=on&sort=popular&target=title&date_range=5y&users=1" >1 user</a></li> <li><a href="/q/SHAP?sort=popular&safe=on&users=3&date_range=5y&target=title" class="is-current">3 users</a></li> <li><a href="/q/SHAP?users=50&date_range=5y&target=title&sort=popular&safe=on" >50 users</a></li> <li><a href="/q/SHAP?target=title&users=100&date_range=5y&safe=on&sort=popular" >100 users</a></li> <li><a href="/q/SHAP?target=title&users=500&date_range=5y&safe=on&sort=popular" >500 users</a></li> </ul> </div> <div class="centerarticle-sidebar-menu js-safe-search-div is-hidden" title="セーフサーチはログイン時にのみoffにできます"> <h3 class="centerarticle-aside-title">セーフサーチ</h3> <ul class="centerarticle-sub-navi"> <li><a href="/q/SHAP?users=3&date_range=5y&target=title&sort=popular&safe=on" class="is-current">オン</a></li> <li><a href="/q/SHAP?target=title&users=3&date_range=5y&safe=off&sort=popular" >オフ</a></li> </ul> </div> <div class="centerarticle-sidebar-menu js-entrysearch-date"> <h3 class="centerarticle-aside-title">期間指定</h3> <ul class="centerarticle-sub-navi"> <li><a href="/q/SHAP?sort=popular&safe=on&date_range=all&users=3&target=title" >すべて</a></li> <li><a href="/q/SHAP?target=title&date_range=w&users=3&safe=on&sort=popular" >1週間</a></li> <li><a href="/q/SHAP?sort=popular&safe=on&users=3&date_range=m&target=title" >1ヶ月</a></li> <li><a href="/q/SHAP?sort=popular&safe=on&users=3&date_range=y&target=title" >1年</a></li> </ul> <form method="get" class="js-entrysearch-datepicker-form"> <input value="SHAP" class="js-entrysearch-form-text-input" type="hidden"> <input value="title" name="target" type="hidden"> <input value="popular" name="sort" type="hidden"> <input value="3" name="users" type="hidden"> <input value="on" name="safe" type="hidden"> <ul class="centerarticle-sub-navi"> <li><input class="js-entrysearch-datepicker-start entrysearch-searchbox-dateInput" value="" type="text" name="date_begin"> から</li> <li><input class="js-entrysearch-datepicker-end entrysearch-searchbox-dateInput" value="" type="text" name="date_end"> まで</li> <li><input value="検索" type="submit" class="entrysearch-searchbox-btn js-entrysearch-form-button"><li> </ul> </form> </div> </div> <div class="entrysearch-bar"> <p class="entrysearch-bar-results">1 - 29 件 / 29件</p> <div class="entrysearch-bar-inner"> <button class="entrysearch-filter-btn js-entrysearch-filter-btn">絞り込み</button> <a class="entrysearch-sort-btn" href="/q/SHAP?date_range=5y&users=3&target=title&sort=popular&safe=on">新着順</a> <a class="entrysearch-sort-btn is-current" href="/q/SHAP?target=title&date_range=5y&users=3&safe=on&sort=recent">人気順</a> </div> </div> <div class="entrysearch-filter js-entrysearch-filter is-hidden"> <div class="entrysearch-filter-header"> <h2 class="entrysearch-filter-title">絞り込み</h2> <button class="entrysearch-filter-close js-entrysearch-filter-close"></button> </div> <ul class="entrysearch-filter-list"> <li class="entrysearch-filter-item js-entrysearch-target-dropdown"> <h3 class="entrysearch-filter-item-head">検索対象</h3> <div class="entrysearch-filter-item-value-wrapper"> <select class="entrysearch-filter-item-value js-entrysearch-filter-target"> <option value="all" >すべて</option> <option value="tag" >タグ</option> <option value="title" selected>タイトル</option> <option value="text" >本文</option> </select> </div> </li> <li class="entrysearch-filter-item"> <h3 class="entrysearch-filter-item-head">ブックマーク数</h3> <div class="entrysearch-filter-item-value-wrapper"> <select class="entrysearch-filter-item-value js-entrysearch-filter-users"> <option value="1" >1 user</option> <option value="3" selected>3 users</option> <option value="50" >50 users</option> <option value="100" >100 users</option> <option value="500" >500 users</option> </select> </div> </li> <li class="entrysearch-filter-item"> <h3 class="entrysearch-filter-item-head">期間</h3> <div class="entrysearch-filter-item-value-wrapper"> <select class="entrysearch-filter-item-value js-entrysearch-filter-daterange"> <option value="all">すべて</option> <option value="w" >1週間</option> <option value="m" >1ヶ月</option> <option value="y" >1年</option> </select> </div> </li> <li class="entrysearch-filter-item js-safe-search-div is-hidden"> <h3 class="entrysearch-filter-item-head">セーフサーチ</h3> <div class="entrysearch-filter-checkbox"> <input type="checkbox" name="safesearch" id="safesearch" class="js-entryseach-safesearch" checked> <label class="entrysearch-filter-checkbox-label" for="safesearch"></label> </div> </li> </ul> <div class="entrysearch-filter-footer"> <button class="entrysearch-filter-search-btn styleguide-btn-primary js-entrysearch-filter-search-btn">検索する</button> <button class="entrysearch-filter-clear-btn styleguide-btn-text js-entrysearch-filter-clear-btn">条件をクリアする</button> </div> </div> <div class="search-container"> <h2 class="entrysearch-title"> <span class="entrysearch-word">SHAPの検索結果</span><span class="entrysearch-result">1 - 29 件 / 29件</span> </h2> <div class="alert alert-gray entrysearch-alert" role="note" data-gtm-inview-label="entry-search-fallback-note"> <p>タグ検索の該当結果が少ないため、タイトル検索結果を表示しています。</p> </div> <div class="entrysearch-summary-text"> <span>SHAP</span>に関するエントリは<span>29</span>件あります。 <span>機械学習</span>、 <span>AI</span>、 <span>Python</span> などが関連タグです。人気エントリには <span>『テキストや画像から3Dモデルを生成するオープンソースのAI「Shap-E」をOpenAIが発表』</span>などがあります。 </div> <div class="entrysearch-related-entries" data-gtm-inview-label="entry-search-recommend-header"> <h3 class="entrysearch-related-title">SHAPの関連エントリー</h3> <ul class="entrysearch-related-list"> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://note.com/elements_hrx/n/ncd2cdaa7a7b1" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/81d4e66a938b6e656ada4c2bd6248bdf762aec58/height=288;version=1;width=512/https%3A%2F%2Fassets.st-note.com%2Fproduction%2Fuploads%2Fimages%2F162529338%2Frectangle_large_type_2_704afdd3422ffa41a3a03eca45d98521.png%3Ffit%3Dbounds%26quality%3D85%26width%3D1280" alt="CEOとCHROに聞いた、創業10年の先に見つけた私たちの新ミッション｜株式会社ELEMENTS"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://note.com/elements_hrx/n/ncd2cdaa7a7b1" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> CEOとCHROに聞いた、創業10年の先に見つけた私たちの新ミッション｜株式会社ELEMENTS </a> </h4> <a href="/entry/s/note.com/elements_hrx/n/ncd2cdaa7a7b1" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 18 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://note.com/elements_hrx/n/n13559ff77e1f" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/35839c4963549f40b10e7e8342347b5a5b14fafb/height=288;version=1;width=512/https%3A%2F%2Fassets.st-note.com%2Fproduction%2Fuploads%2Fimages%2F163285972%2Frectangle_large_type_2_9796e85ef4e9be7a7f991db121d6d33c.png%3Ffit%3Dbounds%26quality%3D85%26width%3D1280" alt="ELEMENTSが7兆円のクラウド市場に挑む理由｜新規事業「ELEMENTS CLOUD」の展望｜株式会社ELEMENTS"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://note.com/elements_hrx/n/n13559ff77e1f" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> ELEMENTSが7兆円のクラウド市場に挑む理由｜新規事業「ELEMENTS CLOUD」の展望｜株式会社ELEMENTS </a> </h4> <a href="/entry/s/note.com/elements_hrx/n/n13559ff77e1f" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 19 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://nafca.jp/public-comment08/" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/27adb35b196090f109d016ca69c6d7f6e6083093/height=288;version=1;width=512/https%3A%2F%2Fnafca.jp%2Fwp-content%2Fuploads%2F2023%2F05%2Fogp.png" alt="パブリックコメント「「生成AIを巡る競争」に関する情報・意見」"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://nafca.jp/public-comment08/" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> パブリックコメント「「生成AIを巡る競争」に関する情報・意見」 </a> </h4> <a href="/entry/s/nafca.jp/public-comment08/" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 23 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://ascii.jp/elem/000/004/236/4236636/" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/b69a4d7b489946a5926201f504b497fbed6b8f4b/height=288;version=1;width=512/https%3A%2F%2Fascii.jp%2Fimg%2F2024%2F11%2F24%2F4123336%2Fxl%2F1508b07bf678c94c.png%3F20200122" alt="3DモデリングにAI革命の兆し　1枚のイラストから3Dデータが完成 (1/4)"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://ascii.jp/elem/000/004/236/4236636/" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> 3DモデリングにAI革命の兆し　1枚のイラストから3Dデータが完成 (1/4) </a> </h4> <a href="/entry/s/ascii.jp/elem/000/004/236/4236636/" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 668 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://nazology.kusuguru.co.jp/archives/166191" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/b992a3fcf515b5dbcfcac3412f53c307bcd24426/height=288;version=1;width=512/https%3A%2F%2Fnazology.kusuguru.co.jp%2Fwp-content%2Fuploads%2F2024%2F11%2F3e6b050622bf0621648ad82326879144.jpg" alt="ほとんどのSNSユーザーはニュースの「タイトル」までしか読まない - ナゾロジー"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://nazology.kusuguru.co.jp/archives/166191" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> ほとんどのSNSユーザーはニュースの「タイトル」までしか読まない - ナゾロジー </a> </h4> <a href="/entry/s/nazology.kusuguru.co.jp/archives/166191" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 14 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://levtech.jp/media/article/focus/detail_565/" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/1ce45a3c5c809bf67717275c4ef6f764291f0b81/height=288;version=1;width=512/https%3A%2F%2Flevtech.jp%2Fmedia%2Fwp-content%2Fuploads%2F2024%2F11%2FLTlab_eyecatch_interview_207.jpg" alt="5指駆動「サイボーグ義手」が実現した理由。“実用化だけ”を見据えた研究者たち【フォーカス】レバテックラボ（レバテックLAB）"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://levtech.jp/media/article/focus/detail_565/" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> 5指駆動「サイボーグ義手」が実現した理由。“実用化だけ”を見据えた研究者たち【フォーカス】レバテックラボ（レバテックLAB） </a> </h4> <a href="/entry/s/levtech.jp/media/article/focus/detail_565/" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 170 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://qiita.com/Isaka-code/items/efbdc60b4c557e2fba1b" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/893b8c454323d3959d42f3f3235769f6e474a1be/height=288;version=1;width=512/https%3A%2F%2Fqiita-user-contents.imgix.net%2Fhttps%253A%252F%252Fqiita-user-contents.imgix.net%252Fhttps%25253A%25252F%25252Fcdn.qiita.com%25252Fassets%25252Fpublic%25252Farticle-ogp-background-afbab5eb44e0b055cce1258705637a91.png%253Fixlib%253Drb-4.0.0%2526w%253D1200%2526blend64%253DaHR0cHM6Ly9xaWl0YS11c2VyLXByb2ZpbGUtaW1hZ2VzLmltZ2l4Lm5ldC9odHRwcyUzQSUyRiUyRnMzLWFwLW5vcnRoZWFzdC0xLmFtYXpvbmF3cy5jb20lMkZxaWl0YS1pbWFnZS1zdG9yZSUyRjAlMkYyNjI3MDE1JTJGZWJmMWRhZjBmMjIyNWUwMDBkOTVlNTc0MjE4NTA2YTYyZDhmMzhlMiUyRnhfbGFyZ2UucG5nJTNGMTY5ODIwNDE0Mj9peGxpYj1yYi00LjAuMCZhcj0xJTNBMSZmaXQ9Y3JvcCZtYXNrPWVsbGlwc2UmZm09cG5nMzImcz1lNGZmZmQ4ZmYxYjhjMWJiN2MzY2U4YmMyYmRkOWJkMQ%2526blend-x%253D120%2526blend-y%253D467%2526blend-w%253D82%2526blend-h%253D82%2526blend-mode%253Dnormal%2526s%253D34666cc930b6299dcc6dc8158d04cde6%3Fixlib%3Drb-4.0.0%26w%3D1200%26fm%3Djpg%26mark64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTk2MCZoPTMyNCZ0eHQ9S2FnZ2xlJUUzJTgxJUE4JUU1JTkwJThEJUUzJTgxJUFFJUUzJTgxJUE0JUUzJTgxJThGJUU2JTlDJUFDJUUzJTgyJTkyJUU1JTg1JUE4JUUzJTgxJUE2JUU4JUFBJUFEJUUzJTgyJTkzJUUzJTgxJUEwJUUzJTgxJUFFJUUzJTgxJUE3JUU3JUI0JUI5JUU0JUJCJThCJUUzJTgxJTk3JUUzJTgxJUE2JUUzJTgxJTg0JUUzJTgxJThGJnR4dC1hbGlnbj1sZWZ0JTJDdG9wJnR4dC1jb2xvcj0lMjMxRTIxMjEmdHh0LWZvbnQ9SGlyYWdpbm8lMjBTYW5zJTIwVzYmdHh0LXNpemU9NTYmdHh0LXBhZD0wJnM9NTNmMzViMWYxZmE0YmViMGZmMTdhZGU2NTlkYzg1M2U%26mark-x%3D120%26mark-y%3D112%26blend64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTgzOCZoPTU4JnR4dD0lNDBJc2FrYS1jb2RlJnR4dC1jb2xvcj0lMjMxRTIxMjEmdHh0LWZvbnQ9SGlyYWdpbm8lMjBTYW5zJTIwVzYmdHh0LXNpemU9MzYmdHh0LXBhZD0wJnM9NThiMDUzMzljY2VhNWM3OTc4N2Q5ZGE0YzZiNDIxYWY%26blend-x%3D242%26blend-y%3D480%26blend-w%3D838%26blend-h%3D46%26blend-fit%3Dcrop%26blend-crop%3Dleft%252Cbottom%26blend-mode%3Dnormal%26s%3D113894b71284c6522636bf664d020b43" alt="Kaggleと名のつく本を全て読んだので紹介していく - Qiita"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://qiita.com/Isaka-code/items/efbdc60b4c557e2fba1b" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> Kaggleと名のつく本を全て読んだので紹介していく - Qiita </a> </h4> <a href="/entry/s/qiita.com/Isaka-code/items/efbdc60b4c557e2fba1b" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 74 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://www.tachibana-akira.com/2024/11/16044" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/common/noimage.png" alt="リベラルの理想世界で最底辺に突き落とされる「やってもできない」ひとたち – 橘玲公式BLOG"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://www.tachibana-akira.com/2024/11/16044" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> リベラルの理想世界で最底辺に突き落とされる「やってもできない」ひとたち – 橘玲公式BLOG </a> </h4> <a href="/entry/s/www.tachibana-akira.com/2024/11/16044" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 17 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://gamemakers.jp/article/2024_11_21_86126/" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/efc20ed63760a5c4a2437019a1f0ae33fa94c879/height=288;version=1;width=512/https%3A%2F%2Fgamemakers.jp%2Fcms%2Fwp-content%2Fuploads%2F2024%2F11%2F15838e3118072fc8668bff05263c03b8.jpg" alt="重さ約2kgのアルゴリズム解説書籍『アルゴリズムイントロダクション第4版総合版』、近代科学社より12/3（火）に発売｜ゲームメーカーズ"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://gamemakers.jp/article/2024_11_21_86126/" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> 重さ約2kgのアルゴリズム解説書籍『アルゴリズムイントロダクション第4版総合版』、近代科学社より12/3（火）に発売｜ゲームメーカーズ </a> </h4> <a href="/entry/s/gamemakers.jp/article/2024_11_21_86126/" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 45 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://github.com/PaulPauls/llama3_interpretability_sae" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/9509619983ef0d45427182d03d706c053f1a5a2e/height=288;version=1;width=512/https%3A%2F%2Fopengraph.githubassets.com%2F193ceacbee92c690801f97f393284e39841e4f340ddb4fc7c430660422aa1ec9%2FPaulPauls%2Fllama3_interpretability_sae" alt="GitHub - PaulPauls/llama3_interpretability_sae: A complete end-to-end pipeline for LLM interpretability with sparse autoencoders (SAEs) using Llama 3.2, written in pure PyTorch and fully reproducible."> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://github.com/PaulPauls/llama3_interpretability_sae" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-header-d-title"> GitHub - PaulPauls/llama3_interpretability_sae: A complete end-to-end pipeline for LLM interpretability with sparse autoencoders (SAEs) using Llama 3.2, written in pure PyTorch and fully reproducible. </a> </h4> <a href="/entry/s/github.com/PaulPauls/llama3_interpretability_sae" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-header-d-users"> 8 users </a> </div> </li> </ul> </div> <ul class="entrysearch-articles"> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://gigazine.net/news/20230510-openai-shap-e/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://gigazine.net/news/20230510-openai-shap-e/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fgigazine.net%2Fnews%2F20230510-openai-shap-e%2F" alt="">テキストや画像から3Dモデルを生成するオープンソースのAI「Shap-E」をOpenAIが発表 </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/gigazine.net/news/20230510-openai-shap-e/" class="js-keyboard-entry-page-openable" title="テキストや画像から3Dモデルを生成するオープンソースのAI「Shap-E」をOpenAIが発表 (65 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 65 users </a> </span> </li> <li> <a href="/site/gigazine.net/" title="『gigazine.net』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> gigazine.net </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/05/10</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">大規模言語モデルのGPT-4やチャットボットAIのChatGPTを開発するOpenAiが、テキストや画像を入力することで3Dモデルを自動生成するAI「Shap-E」を発表しました。Shap-Eはオープンソースで開発されており、無料で使用できます。 GitHub - openai/shap-e: Generate 3D objects conditioned on text or images https://github.com/openai/shap-e 機械学習エンジニアのhysts氏がShap-Eを体験できるデモサイトをAI向けリポジトリサイトのHugging Faceに作成しており、Shap-EがどんなAIなのかをブラウザから簡単に体験できます。 Shap-E - a Hugging Face Space by hysts https://huggingface.co/spaces</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/AI" data-gtm-click-label="entry-search-result-item-tag">AI</a></li> <li><a href="/q/%E4%BA%BA%E5%B7%A5%E7%9F%A5%E8%83%BD" data-gtm-click-label="entry-search-result-item-tag">人工知能</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/OpenAI" data-gtm-click-label="entry-search-result-item-tag">OpenAI</a></li> <li><a href="/q/%E3%83%A2%E3%83%87%E3%83%AB" data-gtm-click-label="entry-search-result-item-tag">モデル</a></li> <li><a href="/q/GitHub" data-gtm-click-label="entry-search-result-item-tag">GitHub</a></li> <li><a href="/q/3D" data-gtm-click-label="entry-search-result-item-tag">3D</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://gigazine.net/news/20230510-openai-shap-e/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://gigazine.net/news/20230510-openai-shap-e/" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/861e6a717ea90641c1811b37af68e0d11fda0c52/height=288;version=1;width=512/https%3A%2F%2Fi.gzn.jp%2Fimg%2F2023%2F05%2F10%2Fopenai-shap-e%2F00.png" alt="テキストや画像から3Dモデルを生成するオープンソースのAI「Shap-E」をOpenAIが発表" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://techblog.jmdc.co.jp/entry/2022/01/11/000000"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://techblog.jmdc.co.jp/entry/2022/01/11/000000" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Ftechblog.jmdc.co.jp%2Fentry%2F2022%2F01%2F11%2F000000" alt="">新型コロナ重症化リスクファクター分析　XGBoost+SHAPによるEDA - JMDC TECH BLOG </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/techblog.jmdc.co.jp/entry/2022/01/11/000000" class="js-keyboard-entry-page-openable" title="新型コロナ重症化リスクファクター分析　XGBoost+SHAPによるEDA - JMDC TECH BLOG (49 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 49 users </a> </span> </li> <li> <a href="/site/techblog.jmdc.co.jp/" title="『techblog.jmdc.co.jp』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> techblog.jmdc.co.jp </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2022/01/11</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">JMDC データサイエンティストの齋藤です。データ分析の第一歩、EDA（探索的データ分析）にどう取り組んでいますか？予測のための機械学習の話はよく聞きますが、EDAのための機械学習はあまり目にしないと感じるので、今回は実務における「XGBoost+SHAPによるEDA」の実践例を取り上げてみたいと思います。題材は２０２１年７月にリリースした「新型コロナウイルス感染時の重症化リスクファクターに関する分析結果」です。 https://www.jmdc.co.jp/wp-content/uploads/2021/07/news20210709_2.pdf このブログの内容はテクニカル中心ですが、分析結果自体も面白いのでレポートもご覧いただけると嬉しいです。 XGBoost+SHAPでEDAする理由分析デザインの概要 Feature Importance SHAP XGBoost+SHA</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E7%B5%B1%E8%A8%88" data-gtm-click-label="entry-search-result-item-tag">統計</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/%E3%83%87%E3%83%BC%E3%82%BF" data-gtm-click-label="entry-search-result-item-tag">データ</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://techblog.jmdc.co.jp/entry/2022/01/11/000000"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://techblog.jmdc.co.jp/entry/2022/01/11/000000" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/9d8d967703fe742177a8d5438743b67db0ed2975/height=288;version=1;width=512/https%3A%2F%2Fcdn-ak.f.st-hatena.com%2Fimages%2Ffotolife%2Ft%2Ftosaito_jmdc%2F20211227%2F20211227170252.png" alt="新型コロナ重症化リスクファクター分析　XGBoost+SHAPによるEDA - JMDC TECH BLOG" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fqiita.com%2Fs1ok69oo%2Fitems%2F0bf92b84e565789a2191" alt="">SHAPで因果関係を説明できる？ - Qiita </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/qiita.com/s1ok69oo/items/0bf92b84e565789a2191" class="js-keyboard-entry-page-openable" title="SHAPで因果関係を説明できる？ - Qiita (47 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 47 users </a> </span> </li> <li> <a href="/site/qiita.com/s1ok69oo" title="『qiita.com/s1ok69oo』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> qiita.com/s1ok69oo </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/02/25</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">はじめに予測モデル（機械学習モデル）を解釈するのに有用なSHAPを用いて因果関係を説明することができるか、についてPythonによるシミュレーションを交えてまとめました。内容に誤り等ございましたら、ご指摘いただけますと幸いです。結論基本的に、SHAPで因果関係は説明できません。これは、SHAPが予測モデルの因果ではなく相関を明らかにするものであるからです。そこで今回は、予測モデルをSHAPで解釈する上でありがちなミスリーディングや、それに関連する因果効果を推定するためのアプローチについて記載しています。そもそもSHAPとは SHAPとはSHapley Additive exPlanationsの略で、協力ゲーム理論のShapley Valueを機械学習に応用した手法です。「その予測モデルがなぜ、その予測値を算出しているか」を解釈するためのツールとしてオープンソースのライブラリが開</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/SHAP" data-gtm-click-label="entry-search-result-item-tag">SHAP</a></li> <li><a href="/q/Python" data-gtm-click-label="entry-search-result-item-tag">Python</a></li> <li><a href="/q/%E5%88%86%E6%9E%90" data-gtm-click-label="entry-search-result-item-tag">分析</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/qiita" data-gtm-click-label="entry-search-result-item-tag">qiita</a></li> <li><a href="/q/%E3%83%87%E3%83%BC%E3%82%BF" data-gtm-click-label="entry-search-result-item-tag">データ</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/afd2b30384de3523c08f78d0dddb66b12555d47e/height=288;version=1;width=512/https%3A%2F%2Fqiita-user-contents.imgix.net%2Fhttps%253A%252F%252Fcdn.qiita.com%252Fassets%252Fpublic%252Farticle-ogp-background-412672c5f0600ab9a64263b751f1bc81.png%3Fixlib%3Drb-4.0.0%26w%3D1200%26mark64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTk3MiZoPTM3OCZ0eHQ9U0hBUCVFMyU4MSVBNyVFNSU5QiVBMCVFNiU5RSU5QyVFOSU5NiVBMiVFNCVCRiU4MiVFMyU4MiU5MiVFOCVBQSVBQyVFNiU5OCU4RSVFMyU4MSVBNyVFMyU4MSU4RCVFMyU4MiU4QiVFRiVCQyU5RiZ0eHQtYWxpZ249bGVmdCUyQ3RvcCZ0eHQtY29sb3I9JTIzMUUyMTIxJnR4dC1mb250PUhpcmFnaW5vJTIwU2FucyUyMFc2JnR4dC1zaXplPTU2JnM9NzExZTZlZTNkZWI5Mzg2Yjk0N2Q5YmM4OTQxMWE1ZGY%26mark-x%3D142%26mark-y%3D57%26blend64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZoPTc2Jnc9NzcwJnR4dD0lNDBzMW9rNjlvbyZ0eHQtY29sb3I9JTIzMUUyMTIxJnR4dC1mb250PUhpcmFnaW5vJTIwU2FucyUyMFc2JnR4dC1zaXplPTM2JnR4dC1hbGlnbj1sZWZ0JTJDdG9wJnM9Y2NkOTI4ZDY0YzBjMzFlY2U0ZWNjNmQxZGRiMGNiNTQ%26blend-x%3D142%26blend-y%3D486%26blend-mode%3Dnormal%26s%3Daafe3a34fafecfbace22fe19baf9ad3a" alt="SHAPで因果関係を説明できる？ - Qiita" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://blog.tsurubee.tech/entry/2021/07/19/120541"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://blog.tsurubee.tech/entry/2021/07/19/120541" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fblog.tsurubee.tech%2Fentry%2F2021%2F07%2F19%2F120541" alt="">協力ゲーム理論のシャープレイ値に基づき機械学習モデルの予測を解釈するKernel SHAPの理論と実装のまとめ - Fire Engine </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/blog.tsurubee.tech/entry/2021/07/19/120541" class="js-keyboard-entry-page-openable" title="協力ゲーム理論のシャープレイ値に基づき機械学習モデルの予測を解釈するKernel SHAPの理論と実装のまとめ - Fire Engine (46 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 46 users </a> </span> </li> <li> <a href="/site/blog.tsurubee.tech/" title="『blog.tsurubee.tech』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> blog.tsurubee.tech </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2021/07/19</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">機械学習の幅広い分野への応用が進むにつれ，機械学習がその予測の根拠などを理解できない「ブラックボックス」となることが問題視されており，機械学習の解釈性や説明性が注目されています．今回のテーマであるSHAP（SHapley Additive exPlanations）は，機械学習モデルへの特定の入力に対する予測の根拠を提示する代表的な手法の一つです．SHAPには用途に応じていくつかのアルゴリズムがありますが，その中でも今回はあらゆる機械学習モデルに適用可能（Model-Agnostic）なKernel SHAPという手法についてまとめました．構成としては，まずKernel SHAPとは何かについての概要を述べた後に， Kernel SHAPを理解する上で必要な要素である「シャープレイ値」と「SHAP」について説明します．さいごに，Kernel SHAPについて「理論」と「実装」に分けて書い</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/SHAP" data-gtm-click-label="entry-search-result-item-tag">SHAP</a></li> <li><a href="/q/interpretability" data-gtm-click-label="entry-search-result-item-tag">interpretability</a></li> <li><a href="/q/machinelearning" data-gtm-click-label="entry-search-result-item-tag">machinelearning</a></li> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/ML" data-gtm-click-label="entry-search-result-item-tag">ML</a></li> <li><a href="/q/Python" data-gtm-click-label="entry-search-result-item-tag">Python</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://blog.tsurubee.tech/entry/2021/07/19/120541"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://blog.tsurubee.tech/entry/2021/07/19/120541" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/a4a9672f8bc153093d20d34e6fe478e0ac067614/height=288;version=1;width=512/https%3A%2F%2Fcdn-ak.f.st-hatena.com%2Fimages%2Ffotolife%2Fh%2Fhirotsuru314%2F20210715%2F20210715215701.png" alt="協力ゲーム理論のシャープレイ値に基づき機械学習モデルの予測を解釈するKernel SHAPの理論と実装のまとめ - Fire Engine" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <div class="related-entry-wrapper"> <div class="entrysearch-related-entries" data-gtm-inview-label="entry-search-recommend-middle"> <h3 class="entrysearch-related-title">SHAPの関連エントリー</h3> <ul class="entrysearch-related-list"> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://www.itmedia.co.jp/news/articles/2411/26/news100.html" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/7034c064d827a8c98717d21521f772a96d890dc6/height=288;version=1;width=512/https%3A%2F%2Fimage.itmedia.co.jp%2Fnews%2Farticles%2F2411%2F26%2Fcover_news100.jpg" alt="XのマスクCTO、「リンクはメインではなくリプライに入れろ」"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://www.itmedia.co.jp/news/articles/2411/26/news100.html" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> XのマスクCTO、「リンクはメインではなくリプライに入れろ」 </a> </h4> <a href="/entry/s/www.itmedia.co.jp/news/articles/2411/26/news100.html" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 338 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://togetter.com/li/2470890" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/1a47d147d76c3b76f7b6f187f35ccae620fbd1ac/height=288;version=1;width=512/https%3A%2F%2Fs.tgstc.com%2Fogp3%2Fd49ae55cf30b3fd3112c965dea75893a-1200x630.jpeg" alt="8歳で国家資格の第2種電気工事士に合格した小学3年生の女の子が話題に→｢技能試験もあったはずなのに圧着ペンチを握れるのか...｣"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://togetter.com/li/2470890" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> 8歳で国家資格の第2種電気工事士に合格した小学3年生の女の子が話題に→｢技能試験もあったはずなのに圧着ペンチを握れるのか...｣ </a> </h4> <a href="/entry/s/togetter.com/li/2470890" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 132 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://docs.aws.amazon.com/ja_jp/AmazonCloudWatch/latest/monitoring/Best_Practice_Recommended_Alarms_AWS_Services.html" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/common/noimage.png" alt="推奨アラーム - Amazon CloudWatch"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://docs.aws.amazon.com/ja_jp/AmazonCloudWatch/latest/monitoring/Best_Practice_Recommended_Alarms_AWS_Services.html" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> 推奨アラーム - Amazon CloudWatch </a> </h4> <a href="/entry/s/docs.aws.amazon.com/ja_jp/AmazonCloudWatch/latest/monitoring/Best_Practice_Recommended_Alarms_AWS_Services.html" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 258 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://zenn.dev/tmknom/articles/readable-github-cicd-book" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/f4270f2b166a3550349b672ca457f04e1f746e95/height=288;version=1;width=512/https%3A%2F%2Fres.cloudinary.com%2Fzenn%2Fimage%2Fupload%2Fs--v8r5fjhu--%2Fc_fit%252Cg_north_west%252Cl_text%3Anotosansjp-medium.otf_55%3A%2525E6%25259B%2525B8%2525E7%2525B1%25258D%2525E3%252580%25258EGitHub%252520CI%25252FCD%2525E5%2525AE%25259F%2525E8%2525B7%2525B5%2525E3%252582%2525AC%2525E3%252582%2525A4%2525E3%252583%252589%2525E3%252580%25258F%2525E3%252582%252592%2525E8%2525AA%2525AD%2525E3%252581%2525BF%2525E3%252582%252584%2525E3%252581%252599%2525E3%252581%25258F%2525E3%252581%252599%2525E3%252582%25258B%2525E6%25258A%252580%2525E8%2525A1%252593%252Cw_1010%252Cx_90%252Cy_100%2Fg_south_west%252Cl_text%3Anotosansjp-medium.otf_37%3Atmknom%252Cx_203%252Cy_121%2Fg_south_west%252Ch_90%252Cl_fetch%3AaHR0cHM6Ly96ZW5uLWRldi5naXRodWIuaW8vZGVmYXVsdC1hdmF0YXJzL2RhcmsvdC5wbmc%3D%252Cr_max%252Cw_90%252Cx_87%252Cy_95%2Fv1627283836%2Fdefault%2Fog-base-w1200-v2.png" alt="書籍『GitHub CI/CD実践ガイド』を読みやすくする技術"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://zenn.dev/tmknom/articles/readable-github-cicd-book" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> 書籍『GitHub CI/CD実践ガイド』を読みやすくする技術 </a> </h4> <a href="/entry/s/zenn.dev/tmknom/articles/readable-github-cicd-book" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 72 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://busmagazine.bestcarweb.jp/news/192941" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/9aae5178c7bd3bb212c01d99f370654a03f45865/height=288;version=1;width=512/https%3A%2F%2Fimg.bestcarweb.jp%2Fwp-content%2Fuploads%2Fsites%2F3%2F2024%2F11%2F18120645%2F5e4896966c9944fcf00890f4502605bf.jpg" alt="熊本のバス５社「共同経営」だと!?　しかしそのデータ分析システムがすごかった！ - バス総合情報誌「バスマガジン」公式WEBサイト"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://busmagazine.bestcarweb.jp/news/192941" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> 熊本のバス５社「共同経営」だと!?　しかしそのデータ分析システムがすごかった！ - バス総合情報誌「バスマガジン」公式WEBサイト </a> </h4> <a href="/entry/s/busmagazine.bestcarweb.jp/news/192941" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 69 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://kaminashi-developer.hatenablog.jp/entry/2024/11/26/080000" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/f7be352aee0b71bfaef5f903d5b740812e0b14a7/height=288;version=1;width=512/https%3A%2F%2Fcdn.image.st-hatena.com%2Fimage%2Fscale%2F9c33c840d1df290a55514ec125738d889a6dcddf%2Fbackend%3Dimagemagick%3Bheight%3D1300%3Bversion%3D1%3Bwidth%3D1300%2Fhttps%253A%252F%252Fcdn-ak.f.st-hatena.com%252Fimages%252Ffotolife%252Fk%252Fkaminashi-developer%252F20241125%252F20241125002230.jpg" alt="Web Application のテストを runn で書いて、開発と価値提供を加速する - カミナシエンジニアブログ"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://kaminashi-developer.hatenablog.jp/entry/2024/11/26/080000" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> Web Application のテストを runn で書いて、開発と価値提供を加速する - カミナシエンジニアブログ </a> </h4> <a href="/entry/s/kaminashi-developer.hatenablog.jp/entry/2024/11/26/080000" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 90 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://www3.nhk.or.jp/news/html/20241126/k10014649641000.html" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/d80f20182f773129fd4725a63df21d2ee77b0b21/height=288;version=1;width=512/https%3A%2F%2Fwww3.nhk.or.jp%2Fnews%2Fhtml%2F20241126%2FK10014649641_2411261053_1126112342_01_02.jpg" alt="アマゾンジャパン出品者に値下げ強いたか公取委立ち入り検査 | NHK"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://www3.nhk.or.jp/news/html/20241126/k10014649641000.html" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> アマゾンジャパン出品者に値下げ強いたか公取委立ち入り検査 | NHK </a> </h4> <a href="/entry/s/www3.nhk.or.jp/news/html/20241126/k10014649641000.html" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 60 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://github.com/vcraftjp/VZEditor" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/38009f195b2ddbae577bdc71ee5bf06e7e939938/height=288;version=1;width=512/https%3A%2F%2Fopengraph.githubassets.com%2F934d035c0e4d668f069e7e8afba98c769248d5d42a4b96fa505538970436b0e6%2Fvcraftjp%2FVZEditor" alt="GitHub - vcraftjp/VZEditor: MS-DOS Text Editor"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://github.com/vcraftjp/VZEditor" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> GitHub - vcraftjp/VZEditor: MS-DOS Text Editor </a> </h4> <a href="/entry/s/github.com/vcraftjp/VZEditor" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 57 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://forest.watch.impress.co.jp/docs/news/1642468.html" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/8620973805a61940faf8515a723c9a62989867b2/height=288;version=1;width=512/https%3A%2F%2Fforest.watch.impress.co.jp%2Fimg%2Fwf%2Flist%2F1642%2F468%2Fimage_top.png" alt="MS-DOS用テキストエディター「VZ Editor」のソースコードが「GitHub」で公開／マニュアルや『PC Power』誌の連載なども。大量のマクロの公開にも期待"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://forest.watch.impress.co.jp/docs/news/1642468.html" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> MS-DOS用テキストエディター「VZ Editor」のソースコードが「GitHub」で公開／マニュアルや『PC Power』誌の連載なども。大量のマクロの公開にも期待 </a> </h4> <a href="/entry/s/forest.watch.impress.co.jp/docs/news/1642468.html" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 54 users </a> </div> </li> <li class="entrysearch-related-listitem"> <div class="entrysearch-related-entry"> <a href="https://xtech.nikkei.com/atcl/nxt/column/18/03009/111900002/" class="entrysearch-related-image" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-image"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/56a43ba93c215f070a36d9145b17d7adfb2c7be7/height=288;version=1;width=512/https%3A%2F%2Fxtech.nikkei.com%2Fatcl%2Fnxt%2Fcolumn%2F18%2F03009%2F111900002%2Ftopm.jpg%3F20220512" alt="セキュリティー投資は売上高の「0.5％」を目指せ、ただし投資しすぎても無意味"> </a> <h4 class="entrysearch-related-entry-title"> <a href="https://xtech.nikkei.com/atcl/nxt/column/18/03009/111900002/" target="_blank" rel="noopener" data-gtm-click-label="entry-search-recommend-middle-title"> セキュリティー投資は売上高の「0.5％」を目指せ、ただし投資しすぎても無意味 </a> </h4> <a href="/entry/s/xtech.nikkei.com/atcl/nxt/column/18/03009/111900002/" class="entrysearch-related-users" data-gtm-click-label="entry-search-recommend-middle-users"> 71 users </a> </div> </li> </ul> </div> </div> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://logmi.jp/tech/articles/322738"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://logmi.jp/tech/articles/322738" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Flogmi.jp%2Ftech%2Farticles%2F322738" alt="">機械学習モデルの予測値を解釈する「SHAP」と協力ゲーム理論の考え方 </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/logmi.jp/tech/articles/322738" class="js-keyboard-entry-page-openable" title="機械学習モデルの予測値を解釈する「SHAP」と協力ゲーム理論の考え方 (40 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 40 users </a> </span> </li> <li> <a href="/site/logmi.jp/" title="『logmi.jp』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> logmi.jp </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2020/04/15</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">機械学習を活用すれば、膨大なデータをもとにして、さまざまな事柄について予測値を導き出すことができます。しかし、実際にその予測値を役立てるためには、なぜそうした予測結果になったのかを理解することが重要です。予測値を解釈するための手法として、協力ゲーム理論を応用したSHAP（SHapley Additive exPlanations）という手法があります。TVISION INSIGHTS株式会社でデータサイエンティストマネージャーを務める森下光之助氏が、SHAPの基本的な考え方と、そのベースとなる協力ゲーム理論について解説します。 SHAP（SHapley Additive exPlanations）で予測結果の貢献度を分解する森下光之助氏（以下、森下）：まずは簡単に自己紹介をさせてください。僕はTVISION INSIGHTS株式会社でデータサイエンティストのマネージャーをやっていて、僕</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/techfeed" data-gtm-click-label="entry-search-result-item-tag">techfeed</a></li> <li><a href="/q/python" data-gtm-click-label="entry-search-result-item-tag">python</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://logmi.jp/tech/articles/322738"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://logmi.jp/tech/articles/322738" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/a568200d1ed19d0dc32c3d7ec0452ed9d99e148c/height=288;version=1;width=512/https%3A%2F%2Fimg.logmi.jp%2Farticle_images%2FMLTiLXerpYiaGK7QgdpvFp.png" alt="機械学習モデルの予測値を解釈する「SHAP」と協力ゲーム理論の考え方" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://cpp-learning.com/interpretable-model/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://cpp-learning.com/interpretable-model/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fcpp-learning.com%2Finterpretable-model%2F" alt="">機械学習モデルの説明性・解釈性について -SHAPによる実践あり-｜はやぶさの技術ノート </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/cpp-learning.com/interpretable-model/" class="js-keyboard-entry-page-openable" title="機械学習モデルの説明性・解釈性について -SHAPによる実践あり-｜はやぶさの技術ノート (34 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 34 users </a> </span> </li> <li> <a href="/site/cpp-learning.com/" title="『cpp-learning.com』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> cpp-learning.com </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2020/05/01</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">機械学習とタスクについて回帰や分類などのタスクに機械学習モデルを活用することがあります。例えば、以下の記事ではフクロウの種類を分類するために深層学習モデルを活用しています。 PyTorch Lightning入門から実践まで -自前データセットで学習し画像分類モデルを生成-ディープラーニングフレームワークPytorchの軽量ラッパー”pytorch-lightning”の入門から実践までのチュートリアル記事を書きました。自前データセットを学習して画像分類モデルを生成し、そのモデルを使って推論するところまでソースコード付で解説しています。...</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/%E8%AA%AC%E6%98%8E%E5%8F%AF%E8%83%BD%E6%80%A7" data-gtm-click-label="entry-search-result-item-tag">説明可能性</a></li> <li><a href="/q/%E8%A7%A3%E9%87%88%E6%80%A7" data-gtm-click-label="entry-search-result-item-tag">解釈性</a></li> <li><a href="/q/LIME" data-gtm-click-label="entry-search-result-item-tag">LIME</a></li> <li><a href="/q/%E3%83%A2%E3%83%87%E3%83%AB" data-gtm-click-label="entry-search-result-item-tag">モデル</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/%E8%AA%AC%E6%98%8E" data-gtm-click-label="entry-search-result-item-tag">説明</a></li> <li><a href="/q/HotEntry" data-gtm-click-label="entry-search-result-item-tag">HotEntry</a></li> <li><a href="/q/Python" data-gtm-click-label="entry-search-result-item-tag">Python</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://cpp-learning.com/interpretable-model/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://cpp-learning.com/interpretable-model/" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/0829dc86893bc98d26fbb7df9543bbb235c150d9/height=288;version=1;width=512/https%3A%2F%2Fcpp-learning.com%2Fwp-content%2Fuploads%2F2020%2F04%2Fmodel-interpretability2.jpg" alt="機械学習モデルの説明性・解釈性について -SHAPによる実践あり-｜はやぶさの技術ノート" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://gigazine.net/news/20230611-shap-e-google-colab/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://gigazine.net/news/20230611-shap-e-google-colab/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fgigazine.net%2Fnews%2F20230611-shap-e-google-colab%2F" alt="">テキスト・画像から3Dモデルを作成するAI「Shap-E」をGoogle Colaboratoryで使ってみた </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/gigazine.net/news/20230611-shap-e-google-colab/" class="js-keyboard-entry-page-openable" title="テキスト・画像から3Dモデルを作成するAI「Shap-E」をGoogle Colaboratoryで使ってみた (31 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 31 users </a> </span> </li> <li> <a href="/site/gigazine.net/" title="『gigazine.net』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> gigazine.net </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/06/11</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">チャットAIのChatGPTや音声認識AIのWhisperを開発するOpenAIが2023年5月に3Dモデル作成AI「Shap-E」を発表しました。Shap-Eはオープンソースで開発されており、誰でも利用可能とのことなので実際にGoogle Colaboratory上で使ってみました。 shap-e/sample_text_to_3d.ipynb at main · openai/shap-e · GitHub https://github.com/openai/shap-e/blob/main/shap_e/examples/sample_text_to_3d.ipynb Shap-Eでどんなことができるのかについては下記の記事が詳しいです。テキストや画像から3Dモデルを生成するオープンソースのAI「Shap-E」をOpenAIが発表 - GIGAZINE まずはGoogleドライブに</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/%2A%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">*あとで読む</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://gigazine.net/news/20230611-shap-e-google-colab/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://gigazine.net/news/20230611-shap-e-google-colab/" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/85ea43e5ea4d7437fb36aed4f2082f3da59d22a8/height=288;version=1;width=512/https%3A%2F%2Fi.gzn.jp%2Fimg%2F2023%2F06%2F11%2Fshap-e-google-colab%2F00.png" alt="テキスト・画像から3Dモデルを作成するAI「Shap-E」をGoogle Colaboratoryで使ってみた" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://www.datarobot.com/jp/blog/explain-machine-learning-models-using-shap/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://www.datarobot.com/jp/blog/explain-machine-learning-models-using-shap/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fwww.datarobot.com%2Fjp%2Fblog%2Fexplain-machine-learning-models-using-shap%2F" alt="">SHAPを用いて機械学習モデルを説明する l DataRobot </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/www.datarobot.com/jp/blog/explain-machine-learning-models-using-shap/" class="js-keyboard-entry-page-openable" title="SHAPを用いて機械学習モデルを説明する l DataRobot (31 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 31 users </a> </span> </li> <li> <a href="/site/www.datarobot.com/" title="『www.datarobot.com』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> www.datarobot.com </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2021/04/14</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">プラットフォームの概要 AI Platform 生成 AIおよび予測 AIのプラットフォームもっと詳しくドキュメント新機能ログイン無料で始める運用自信を持ってAIを拡張し、比類のないエンタープライズ・モニタリングとコントロールでビジネス価値を促進デプロイと実行再学習と最適化監視と介入ガバナンス AIの環境、チーム、およびワークフローを統合し、大規模な範囲での完全な可視性と監視を実現レジストリと管理監査と承認コンプライアンスドキュメント生成構築ニーズの進化に合わせて自由に適応できるオープンなAIエコシステムで、迅速なイノベーションを実現分析と変換学習とチューニング組立てと比較プラットフォーム統合インフラストラクチャーへのデプロイソリューション業界ごとヘルスケア製造小売業金融サービス成果ごとユースケースのライブラリーお客様事例 Dat</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/SHAP" data-gtm-click-label="entry-search-result-item-tag">SHAP</a></li> <li><a href="/q/%E8%AA%AC%E6%98%8E%E5%8F%AF%E8%83%BD%E6%80%A7" data-gtm-click-label="entry-search-result-item-tag">説明可能性</a></li> <li><a href="/q/MachineLearning" data-gtm-click-label="entry-search-result-item-tag">MachineLearning</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/%E3%83%8D%E3%82%BF" data-gtm-click-label="entry-search-result-item-tag">ネタ</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://www.datarobot.com/jp/blog/explain-machine-learning-models-using-shap/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://www.datarobot.com/jp/blog/explain-machine-learning-models-using-shap/" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/33d7a4b0a03fbcfb3d8af98d2b0157cb1c919091/height=288;version=1;width=512/https%3A%2F%2Fwww.datarobot.com%2Fjp%2Fwp-content%2Fuploads%2Fsites%2F2%2F2021%2F03%2F101-Shap_Blog.png" alt="SHAPを用いて機械学習モデルを説明する l DataRobot" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://speakerdeck.com/smorce/shap-lime-pdp-grad-cam"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://speakerdeck.com/smorce/shap-lime-pdp-grad-cam" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fspeakerdeck.com%2Fsmorce%2Fshap-lime-pdp-grad-cam" alt="">SHAPやLIMEなどの説明AI（XAI）のご紹介 / SHAP LIME PDP Grad-CAM </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/speakerdeck.com/smorce/shap-lime-pdp-grad-cam" class="js-keyboard-entry-page-openable" title="SHAPやLIMEなどの説明AI（XAI）のご紹介 / SHAP LIME PDP Grad-CAM (27 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 27 users </a> </span> </li> <li> <a href="/site/speakerdeck.com/smorce" title="『speakerdeck.com/smorce』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> speakerdeck.com/smorce </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/11/10</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">************************************************************* # 要約 ************************************************************* ・SHAP のアルゴリズムを説明・SHAPに近い手法「LIME」との比較及び考察も記載・SHAP 以外の手法で算出した Feature Importance との比較及び考察も記載・オープンソースのデータ及びコードを使ってハンズオン ┗テーブルデータ and 画像データに適用・SHAP 以外の画像説明モデル Grad-CAM, Guided Grad-CAM の紹介</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/AI" data-gtm-click-label="entry-search-result-item-tag">AI</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://speakerdeck.com/smorce/shap-lime-pdp-grad-cam"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://speakerdeck.com/smorce/shap-lime-pdp-grad-cam" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/1c9485dac85b02419c664c24f5d0ee20840c6144/height=288;version=1;width=512/https%3A%2F%2Ffiles.speakerdeck.com%2Fpresentations%2Fdbcb611f9e0e4598ab65e064fd72daef%2Fslide_0.jpg%3F18829709" alt="SHAPやLIMEなどの説明AI（XAI）のご紹介 / SHAP LIME PDP Grad-CAM" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://qiita.com/7mpy/items/c22a0312939fc9a46099"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://qiita.com/7mpy/items/c22a0312939fc9a46099" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fqiita.com%2F7mpy%2Fitems%2Fc22a0312939fc9a46099" alt="">テキスト/画像から3Dモデルを生成する【Shap-E】 - Qiita </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/qiita.com/7mpy/items/c22a0312939fc9a46099" class="js-keyboard-entry-page-openable" title="テキスト/画像から3Dモデルを生成する【Shap-E】 - Qiita (25 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 25 users </a> </span> </li> <li> <a href="/site/qiita.com/7mpy" title="『qiita.com/7mpy』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> qiita.com/7mpy </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/05/05</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">使い方pip install -e . でインストール上記のようなモデルを生成するには、次のノートブックを参照してください。: sample_text_to_3d.ipynb - テキストプロンプトで条件付けされた 3D モデルのサンプル sample_image_to_3d.ipynb - 合成ビュー画像で調整された 3D モデルをサンプリングします encode_model.ipynb - 3D モデルまたはトリメッシュをロードし、マルチビューレンダリングとポイントクラウドのバッチを作成し、それらを潜在的にエンコードして、レンダリングし直します。これを機能させるには、Blender バージョン 3.3.1 以降をインストールし、環境変数BLENDER_PATHを Blender 実行可能ファイルのパスに設定します。テキストから3Dモデルを生成するimport torch fr</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/3DCG" data-gtm-click-label="entry-search-result-item-tag">3DCG</a></li> <li><a href="/q/AI" data-gtm-click-label="entry-search-result-item-tag">AI</a></li> <li><a href="/q/%E3%83%A2%E3%83%87%E3%83%AB" data-gtm-click-label="entry-search-result-item-tag">モデル</a></li> <li><a href="/q/qiita" data-gtm-click-label="entry-search-result-item-tag">qiita</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/3D" data-gtm-click-label="entry-search-result-item-tag">3D</a></li> <li><a href="/q/%E7%94%BB%E5%83%8F" data-gtm-click-label="entry-search-result-item-tag">画像</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://qiita.com/7mpy/items/c22a0312939fc9a46099"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://qiita.com/7mpy/items/c22a0312939fc9a46099" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/2a9d718b95020bc88a2d168cc65b15ae155db8d9/height=288;version=1;width=512/https%3A%2F%2Fqiita-user-contents.imgix.net%2Fhttps%253A%252F%252Fcdn.qiita.com%252Fassets%252Fpublic%252Farticle-ogp-background-412672c5f0600ab9a64263b751f1bc81.png%3Fixlib%3Drb-4.0.0%26w%3D1200%26mark64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTk3MiZoPTM3OCZ0eHQ9JUUzJTgzJTg2JUUzJTgyJUFEJUUzJTgyJUI5JUUzJTgzJTg4JTJGJUU3JTk0JUJCJUU1JTgzJThGJUUzJTgxJThCJUUzJTgyJTg5M0QlRTMlODMlQTIlRTMlODMlODclRTMlODMlQUIlRTMlODIlOTIlRTclOTQlOUYlRTYlODglOTAlRTMlODElOTklRTMlODIlOEIlRTMlODAlOTBTaGFwLUUlRTMlODAlOTEmdHh0LWFsaWduPWxlZnQlMkN0b3AmdHh0LWNvbG9yPSUyMzFFMjEyMSZ0eHQtZm9udD1IaXJhZ2lubyUyMFNhbnMlMjBXNiZ0eHQtc2l6ZT01NiZzPWMyNzE3ZDIwNWIzZjE4ODM4Njk2YTQxYmY1MDE1N2Fi%26mark-x%3D142%26mark-y%3D57%26blend64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZoPTc2Jnc9NzcwJnR4dD0lNDA3bXB5JnR4dC1jb2xvcj0lMjMxRTIxMjEmdHh0LWZvbnQ9SGlyYWdpbm8lMjBTYW5zJTIwVzYmdHh0LXNpemU9MzYmdHh0LWFsaWduPWxlZnQlMkN0b3Amcz1lOTNhMTJkYTJiMWU0Zjc1NzQ2YmQ2YWQ4Y2ExNjY5ZQ%26blend-x%3D142%26blend-y%3D486%26blend-mode%3Dnormal%26s%3De685a8a3e8f94801e7cff0dfec36cc8e" alt="テキスト/画像から3Dモデルを生成する【Shap-E】 - Qiita" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://techblog.gmo-ap.jp/2022/12/20/shap_nlp_negapoji/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://techblog.gmo-ap.jp/2022/12/20/shap_nlp_negapoji/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Ftechblog.gmo-ap.jp%2F2022%2F12%2F20%2Fshap_nlp_negapoji%2F" alt="">SHAPで自然言語処理モデルネガポジ判定の中身を覗いてみる </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/techblog.gmo-ap.jp/2022/12/20/shap_nlp_negapoji/" class="js-keyboard-entry-page-openable" title="SHAPで自然言語処理モデルネガポジ判定の中身を覗いてみる (16 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 16 users </a> </span> </li> <li> <a href="/site/techblog.gmo-ap.jp/" title="『techblog.gmo-ap.jp』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> techblog.gmo-ap.jp </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2022/12/20</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">この記事は GMOアドマーケティングAdvent Calendar 2022 20日目の記事です。みなさんこんにちは、GMOアドマーケティングのM.Hです。今回はXAIの一つである「SHAP」というライブラリについて、自然言語処理向けの機械学習モデルの観点から書いていこうと思います。はじめに昨今ではもはや聞き馴染みとなった「機械学習」や「AI」ですが、そのモデルはコンピューティングシステムの計算能力の向上と共に加速度的な速さで複雑化してきています。もちろんそのようなモデルを使ってビジネスに貢献ができれば嬉しいのですが、モデルの中身に関してはブラックボックス化されていることがほとんどで、「何が要因となってこの結果がもたらされたのか？」という部分はなおざりになりがちです。機械学習のビジネス利用が当たり前に行われるようになった今日では、こういった原因や要因に関して人間がわかる形で示そう</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/NLP" data-gtm-click-label="entry-search-result-item-tag">NLP</a></li> <li><a href="/q/Python" data-gtm-click-label="entry-search-result-item-tag">Python</a></li> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/%E3%83%97%E3%83%AD%E3%82%B0%E3%83%A9%E3%83%9F%E3%83%B3%E3%82%B0" data-gtm-click-label="entry-search-result-item-tag">プログラミング</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://techblog.gmo-ap.jp/2022/12/20/shap_nlp_negapoji/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://techblog.gmo-ap.jp/2022/12/20/shap_nlp_negapoji/" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/79cd69eda696478b644bb14a0bdf93f6008ad497/height=288;version=1;width=512/https%3A%2F%2Ftechblog.gmo-ap.jp%2Fwp-content%2Fuploads%2F2022%2F11%2Fefe873b644635cdd289dc0c4379b8978.png" alt="SHAPで自然言語処理モデルネガポジ判定の中身を覗いてみる" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://github.com/openai/shap-e"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://github.com/openai/shap-e" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fgithub.com%2Fopenai%2Fshap-e" alt="">GitHub - openai/shap-e: Generate 3D objects conditioned on text or images </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/github.com/openai/shap-e" class="js-keyboard-entry-page-openable" title="GitHub - openai/shap-e: Generate 3D objects conditioned on text or images (16 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 16 users </a> </span> </li> <li> <a href="/site/github.com/openai" title="『github.com/openai』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> github.com/openai </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/05/06</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session. Dismiss alert</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/AI" data-gtm-click-label="entry-search-result-item-tag">AI</a></li> <li><a href="/q/3D" data-gtm-click-label="entry-search-result-item-tag">3D</a></li> <li><a href="/q/github" data-gtm-click-label="entry-search-result-item-tag">github</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://github.com/openai/shap-e"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://github.com/openai/shap-e" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/1e1a88146970e5aa88df2f419491c326181882db/height=288;version=1;width=512/https%3A%2F%2Fopengraph.githubassets.com%2Fca467ac0e5526d1cd558e6dacc5be2b53f7905ec2b2571fd03e52cb909a39d41%2Fopenai%2Fshap-e" alt="GitHub - openai/shap-e: Generate 3D objects conditioned on text or images" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://blog.tsurubee.tech/entry/2021/07/21/094213"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://blog.tsurubee.tech/entry/2021/07/21/094213" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fblog.tsurubee.tech%2Fentry%2F2021%2F07%2F21%2F094213" alt="">機械学習の予測を解釈するKernel SHAPの高速性と拡張性の向上を目指したライブラリを開発した - Fire Engine </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/blog.tsurubee.tech/entry/2021/07/21/094213" class="js-keyboard-entry-page-openable" title="機械学習の予測を解釈するKernel SHAPの高速性と拡張性の向上を目指したライブラリを開発した - Fire Engine (15 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 15 users </a> </span> </li> <li> <a href="/site/blog.tsurubee.tech/" title="『blog.tsurubee.tech』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> blog.tsurubee.tech </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2021/07/21</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">先日，協力ゲーム理論のシャープレイ値に基づき機械学習モデルの予測を解釈するKernel SHAPという手法の理論と既存のライブラリの実装についてのブログを書いた． blog.tsurubee.tech 既存のSHAPライブラリであるslundberg/shap（以下，単にSHAPライブラリと呼ぶ）は，SHAPの提案論文*1のファーストオーサーにより開発されており，多くのSHAPのアルゴリズムの実装や可視化の機能が非常に充実している素晴らしいライブラリである．しかし，私が自身の研究の中でSHAPライブラリの中のKernel SHAPを使っている際に，計算速度と拡張のしやすさの観点で改善したいポイントがいくつか出てきた．今回は，まだ絶賛開発中であるが，Kernel SHAPの高速性と拡張性の向上を目指したShapPackというライブラリのプロトタイプが完成したので，それについて紹介する．目</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/python" data-gtm-click-label="entry-search-result-item-tag">python</a></li> <li><a href="/q/%E5%8B%89%E5%BC%B7" data-gtm-click-label="entry-search-result-item-tag">勉強</a></li> <li><a href="/q/AI" data-gtm-click-label="entry-search-result-item-tag">AI</a></li> <li><a href="/q/%E3%83%97%E3%83%AD%E3%82%B0%E3%83%A9%E3%83%9F%E3%83%B3%E3%82%B0" data-gtm-click-label="entry-search-result-item-tag">プログラミング</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://blog.tsurubee.tech/entry/2021/07/21/094213"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://blog.tsurubee.tech/entry/2021/07/21/094213" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/608cc611f315cc0f631e536db1cefd4db69bb095/height=288;version=1;width=512/https%3A%2F%2Fcdn-ak.f.st-hatena.com%2Fimages%2Ffotolife%2Fh%2Fhirotsuru314%2F20210720%2F20210720165023.png" alt="機械学習の予測を解釈するKernel SHAPの高速性と拡張性の向上を目指したライブラリを開発した - Fire Engine" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://huggingface.co/spaces/hysts/Shap-E"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://huggingface.co/spaces/hysts/Shap-E" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fhuggingface.co%2Fspaces%2Fhysts%2FShap-E" alt="">Shap-E - a Hugging Face Space by hysts </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/huggingface.co/spaces/hysts/Shap-E" class="js-keyboard-entry-page-openable" title="Shap-E - a Hugging Face Space by hysts (13 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 13 users </a> </span> </li> <li> <a href="/site/huggingface.co/" title="『huggingface.co』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> huggingface.co </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/05/09</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">text-to-3D & image-to-3D</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/AI" data-gtm-click-label="entry-search-result-item-tag">AI</a></li> <li><a href="/q/3dcg" data-gtm-click-label="entry-search-result-item-tag">3dcg</a></li> <li><a href="/q/3D" data-gtm-click-label="entry-search-result-item-tag">3D</a></li> <li><a href="/q/%E6%96%87%E7%AB%A0" data-gtm-click-label="entry-search-result-item-tag">文章</a></li> <li><a href="/q/%E3%83%86%E3%82%AF%E3%83%8E%E3%83%AD%E3%82%B8%E3%83%BC" data-gtm-click-label="entry-search-result-item-tag">テクノロジー</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://huggingface.co/spaces/hysts/Shap-E"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://huggingface.co/spaces/hysts/Shap-E" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/5659e239a73d87d37c3546825141e4e8756789a1/height=288;version=1;width=512/https%3A%2F%2Fcdn-thumbnails.huggingface.co%2Fsocial-thumbnails%2Fspaces%2Fhysts%2FShap-E.png" alt="Shap-E - a Hugging Face Space by hysts" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://ledge.ai/articles/openai_shap-e"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://ledge.ai/articles/openai_shap-e" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fledge.ai%2Farticles%2Fopenai_shap-e" alt="">OpenAIが3Dモデル自動生成AI「Shap-E」を発表 | Ledge.ai </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/ledge.ai/articles/openai_shap-e" class="js-keyboard-entry-page-openable" title="OpenAIが3Dモデル自動生成AI「Shap-E」を発表 | Ledge.ai (13 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 13 users </a> </span> </li> <li> <a href="/site/ledge.ai/" title="『ledge.ai』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> ledge.ai </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/05/23</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">サインインした状態で「いいね」を押すと、マイページの「いいね履歴」に一覧として保存されていくので、再度読みたくなった時や、あとでじっくり読みたいときに便利です。</p> <ul class="entrysearch-entry-tags"> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://ledge.ai/articles/openai_shap-e"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://ledge.ai/articles/openai_shap-e" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/12564bc6a29c4854f0bea5e21c20a3bae5a11822/height=288;version=1;width=512/https%3A%2F%2Fstorage.googleapis.com%2Fledge-ai-prd-public-bucket%2Fmedia%2Fshape_764abc4131%2Fshape_764abc4131.jpeg" alt="OpenAIが3Dモデル自動生成AI「Shap-E」を発表 | Ledge.ai" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://xtech.nikkei.com/atcl/nxt/column/18/01147/122000002/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://xtech.nikkei.com/atcl/nxt/column/18/01147/122000002/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fxtech.nikkei.com%2Fatcl%2Fnxt%2Fcolumn%2F18%2F01147%2F122000002%2F" alt="">見えてきた説明可能AIの「限界」、代表格LIMEとSHAPの最新動向を知る </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/xtech.nikkei.com/atcl/nxt/column/18/01147/122000002/" class="js-keyboard-entry-page-openable" title="見えてきた説明可能AIの「限界」、代表格LIMEとSHAPの最新動向を知る (9 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 9 users </a> </span> </li> <li> <a href="/site/xtech.nikkei.com/" title="『xtech.nikkei.com』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> xtech.nikkei.com </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2019/12/24</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">「ここ2年ほど、主要な国際AI学会では新しい『説明可能なAI』の手法の提案よりも、どの説明可能AIが実社会で使えるか、生成する説明に信頼性があるか、などを検証する研究が増えてきた」。説明可能AIの研究動向に詳しい大阪大学産業科学研究所の原聡助教はこう語る。説明可能AIへのニーズや実用化への機運が産業界で高まりつつある一方、大学など学術界においては技術の限界や過信へのリスクに警鐘を鳴らす研究が目立ってきた。説明可能AIの実用化に向けた、産業界と学術界双方の動きを追った。説明可能AIの代表選手「LIME」と「SHAP」説明可能AIの手法はいくつかあるが、特に産業界において代表的な手法と目されているのが「LIME（local interpretable model-agnostic explanations）」と、その発展形である「SHAP（SHapley Additive exPlan</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/HotEntry" data-gtm-click-label="entry-search-result-item-tag">HotEntry</a></li> <li><a href="/q/%E3%81%82%E3%81%A8%E3%81%A7%E8%AA%AD%E3%82%80" data-gtm-click-label="entry-search-result-item-tag">あとで読む</a></li> <li><a href="/q/%E5%A4%A7%E5%AD%A6" data-gtm-click-label="entry-search-result-item-tag">大学</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://xtech.nikkei.com/atcl/nxt/column/18/01147/122000002/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://xtech.nikkei.com/atcl/nxt/column/18/01147/122000002/" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/0640b349a335c23e420b77175639674f9a07e516/height=288;version=1;width=512/https%3A%2F%2Fxtech.nikkei.com%2Fatcl%2Fnxt%2Fcolumn%2F18%2F01147%2F122000002%2Ftopm.jpg%3F20220512" alt="見えてきた説明可能AIの「限界」、代表格LIMEとSHAPの最新動向を知る" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://blog.amedama.jp/entry/shap-lightgbm"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://blog.amedama.jp/entry/shap-lightgbm" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fblog.amedama.jp%2Fentry%2Fshap-lightgbm" alt="">Python: SHAP (SHapley Additive exPlanations) を LightGBM と使ってみる - CUBE SUGAR CONTAINER </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/blog.amedama.jp/entry/shap-lightgbm" class="js-keyboard-entry-page-openable" title="Python: SHAP (SHapley Additive exPlanations) を LightGBM と使ってみる - CUBE SUGAR CONTAINER (7 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 7 users </a> </span> </li> <li> <a href="/site/blog.amedama.jp/" title="『blog.amedama.jp』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> blog.amedama.jp </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2020/08/13</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">SHAP は協力ゲーム理論にもとづいて機械学習モデルを解釈する手法と、その実装を指している。今回は、あまり理論の部分には踏み込むことなく、使い方を中心として書いていく。使った環境は次のとおり。 $ sw_vers ProductName: Mac OS X ProductVersion: 10.15.6 BuildVersion: 19G73 $ python -V Python 3.8.5 下準備はじめに、利用するパッケージをインストールしておく。 $ pip install shap lightgbm scikit-learn matplotlib jupyterlab また、SHAP は Jupyter 経由で使った方がインタラクティブな表示ができる。そのため、今回は Jupyter Lab 上で操作することを想定する。 $ jupyter-lab Jupyter Lab を</p> <ul class="entrysearch-entry-tags"> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://blog.amedama.jp/entry/shap-lightgbm"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://blog.amedama.jp/entry/shap-lightgbm" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/6b9967312d325f2d32b47572f99d9d4a6f811b4f/height=288;version=1;width=512/https%3A%2F%2Fcdn-ak.f.st-hatena.com%2Fimages%2Ffotolife%2Fm%2Fmomijiame%2F20200813%2F20200813184109.png" alt="Python: SHAP (SHapley Additive exPlanations) を LightGBM と使ってみる - CUBE SUGAR CONTAINER" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://speakerdeck.com/s1ok69oo/shapteyin-guo-guan-xi-woshuo-ming-tekiru"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://speakerdeck.com/s1ok69oo/shapteyin-guo-guan-xi-woshuo-ming-tekiru" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fspeakerdeck.com%2Fs1ok69oo%2Fshapteyin-guo-guan-xi-woshuo-ming-tekiru" alt="">SHAPで因果関係を説明できる？ </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/speakerdeck.com/s1ok69oo/shapteyin-guo-guan-xi-woshuo-ming-tekiru" class="js-keyboard-entry-page-openable" title="SHAPで因果関係を説明できる？ (6 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 6 users </a> </span> </li> <li> <a href="/site/speakerdeck.com/s1ok69oo" title="『speakerdeck.com/s1ok69oo』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> speakerdeck.com/s1ok69oo </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/06/19</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">2023年6月19日に社内で発表したLT資料です． SHAPでありがちなミスリーディングや，特徴量が予測に与える影響（因果効果）を推定できるケースについて紹介しました．資料内で出てくるリンクは以下です（参照日: 2023年6月13日） https://arxiv.org/abs/2011.01625 https://christophm.github.io/interpretable-ml-book/shapley.html https://christophm.github.io/interpretable-ml-book/shap.html https://econml.azurewebsites.net/ https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191 https://shap.readthedocs.io/en/late</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E7%B5%B1%E8%A8%88" data-gtm-click-label="entry-search-result-item-tag">統計</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://speakerdeck.com/s1ok69oo/shapteyin-guo-guan-xi-woshuo-ming-tekiru"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://speakerdeck.com/s1ok69oo/shapteyin-guo-guan-xi-woshuo-ming-tekiru" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/0a33d6b3fab9f29680a39f4c24842d42281d408e/height=288;version=1;width=512/https%3A%2F%2Ffiles.speakerdeck.com%2Fpresentations%2F557f5b089cf74deaa75a7dbf01c2570b%2Fslide_0.jpg%3F25985110" alt="SHAPで因果関係を説明できる？" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://rightcode.co.jp/blog/information-technology/shap-tree-structure-model-interpretation"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://rightcode.co.jp/blog/information-technology/shap-tree-structure-model-interpretation" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Frightcode.co.jp%2Fblog%2Finformation-technology%2Fshap-tree-structure-model-interpretation" alt="">【機械学習】SHAPを用いた木構造モデルの解釈 – 株式会社ライトコード </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/rightcode.co.jp/blog/information-technology/shap-tree-structure-model-interpretation" class="js-keyboard-entry-page-openable" title="【機械学習】SHAPを用いた木構造モデルの解釈 – 株式会社ライトコード (4 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 4 users </a> </span> </li> <li> <a href="/site/rightcode.co.jp/" title="『rightcode.co.jp』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> rightcode.co.jp </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2020/07/07</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">SHAPを用いた木構造モデルの解釈機械学習モデルの解釈性については、しばしば問題になります。「モデルの精度が高いが、なぜモデルが、その予測を行ったのかを説明できず、実用に至れなかった…。」といった事を、多くの方が経験しているのではないでしょうか？今回は、複雑なモデルの解釈を行うための手法「SHAP（SHapley Additive exPlanations）」を紹介します。そして、実際に「木構造モデル」に適用したいと思います。この記事で紹介すること複雑なモデルにおいて、各特徴量の寄与の解釈を行うモデル「SHAP」を解説。「SHAP」を、実際に「木構造モデル」に適用してみた結果を、特徴量重要度の指標と比較する。モデルの解釈性についてまず、機械学習モデルの解釈性について説明します。一般的に、「モデルの複雑性」と「解釈性」は、トレードオフの関係にあります。例えば、「重回帰モデル」</p> <ul class="entrysearch-entry-tags"> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://rightcode.co.jp/blog/information-technology/shap-tree-structure-model-interpretation"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://rightcode.co.jp/blog/information-technology/shap-tree-structure-model-interpretation" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/3d09cc9580513e9769ae008c8f835ffeb2a5bc63/height=288;version=1;width=512/https%3A%2F%2Frightcode.co.jp%2Fwp-content%2Fuploads%2F2020%2F03%2FSHAP.png" alt="【機械学習】SHAPを用いた木構造モデルの解釈 – 株式会社ライトコード" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://qiita.com/ku_a_i/items/74241c84cda8c7327277"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://qiita.com/ku_a_i/items/74241c84cda8c7327277" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fqiita.com%2Fku_a_i%2Fitems%2F74241c84cda8c7327277" alt="">LightGBMの出力結果を解析したい！(SHAPのススメ) - Qiita </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/qiita.com/ku_a_i/items/74241c84cda8c7327277" class="js-keyboard-entry-page-openable" title="LightGBMの出力結果を解析したい！(SHAPのススメ) - Qiita (4 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 4 users </a> </span> </li> <li> <a href="/site/qiita.com/ku_a_i" title="『qiita.com/ku_a_i』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> qiita.com/ku_a_i </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2021/02/09</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">1.はじめに SIGNATEで「日本取引所グループファンダメンタルズ分析チャレンジ」というコンペが開催されています。私も参加していますが、その中で出てくる知識に関して基礎部分をまとめよう！という動機で書いた記事の第2弾です。コンペのチュートリアルでは、「XGBoost」での予測結果に関しての根拠説明として本記事で紹介する「SHAP」を使用していますが、本記事では現在一番メジャーなLightGBMを使用してこのSHAPを紹介していきます。動作環境 OS : Windows10 pro Python : 3.8.3 // Miniconda 4.9.1 LightGBM: 3.1.1 SHAP : 0.38.1 jupyter notebook 2.LightGBMでとりあえず学習して、特徴量重要度を図示する今回は、回帰分析を例として行うことにする。サンプルデータはscikit-le</p> <ul class="entrysearch-entry-tags"> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://qiita.com/ku_a_i/items/74241c84cda8c7327277"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://qiita.com/ku_a_i/items/74241c84cda8c7327277" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/b6ba69e9c5dee22aa8237e369e18c071b463902b/height=288;version=1;width=512/https%3A%2F%2Fqiita-user-contents.imgix.net%2Fhttps%253A%252F%252Fcdn.qiita.com%252Fassets%252Fpublic%252Farticle-ogp-background-412672c5f0600ab9a64263b751f1bc81.png%3Fixlib%3Drb-4.0.0%26w%3D1200%26mark64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTk3MiZoPTM3OCZ0eHQ9TGlnaHRHQk0lRTMlODElQUUlRTUlODclQkElRTUlOEElOUIlRTclQjUlOTAlRTYlOUUlOUMlRTMlODIlOTIlRTglQTclQTMlRTYlOUUlOTAlRTMlODElOTclRTMlODElOUYlRTMlODElODQlRUYlQkMlODElMjhTSEFQJUUzJTgxJUFFJUUzJTgyJUI5JUUzJTgyJUI5JUUzJTgzJUExJTI5JnR4dC1jb2xvcj0lMjMyMTIxMjEmdHh0LWZvbnQ9SGlyYWdpbm8lMjBTYW5zJTIwVzYmdHh0LXNpemU9NTYmdHh0LWFsaWduPWxlZnQlMkN0b3Amcz0wMmQ1MTBiYTQ4YTA4NDRlMWMyM2JhMzJmNzEyOGJhNg%26mark-x%3D142%26mark-y%3D57%26blend64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZoPTc2Jnc9NzcwJnR4dD0lNDBrdV9hX2kmdHh0LWNvbG9yPSUyMzIxMjEyMSZ0eHQtZm9udD1IaXJhZ2lubyUyMFNhbnMlMjBXNiZ0eHQtc2l6ZT0zNiZ0eHQtYWxpZ249bGVmdCUyQ3RvcCZzPWQ2NzJmMzM1MDUyYWY5YjhhN2JjNzQ5NDY3ODJiN2Uz%26blend-x%3D142%26blend-y%3D486%26blend-mode%3Dnormal%26s%3D0212c1135521bb00da181e6b92de93bb" alt="LightGBMの出力結果を解析したい！(SHAPのススメ) - Qiita" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry "> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://research.miidas.jp/2019/11/shap%E5%80%A4%E3%81%A7%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8B%E5%89%8D%E3%81%ABpermutation-importance%E3%82%92%E7%9F%A5%E3%82%8B/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://research.miidas.jp/2019/11/shap%E5%80%A4%E3%81%A7%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8B%E5%89%8D%E3%81%ABpermutation-importance%E3%82%92%E7%9F%A5%E3%82%8B/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fresearch.miidas.jp%2F2019%2F11%2Fshap%25E5%2580%25A4%25E3%2581%25A7%25E8%25A7%25A3%25E9%2587%2588%25E3%2581%2599%25E3%2582%258B%25E5%2589%258D%25E3%2581%25ABpermutation-importance%25E3%2582%2592%25E7%259F%25A5%25E3%2582%258B%2F" alt="">SHAP値で解釈する前にPermutation ImportanceとPDPを知る – MIIDAS Science Blog </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/research.miidas.jp/2019/11/shap%E5%80%A4%E3%81%A7%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8B%E5%89%8D%E3%81%ABpermutation-importance%E3%82%92%E7%9F%A5%E3%82%8B/" class="js-keyboard-entry-page-openable" title="SHAP値で解釈する前にPermutation ImportanceとPDPを知る – MIIDAS Science Blog (4 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 4 users </a> </span> </li> <li> <a href="/site/research.miidas.jp/" title="『research.miidas.jp』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> research.miidas.jp </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2019/12/11</span></li> </ul> </div> <div class="centerarticle-entry-contents "> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">https://www.kaggle.com/learn/machine-learning-explainability 今回はこれをやっていきます。先に一言で言うと機械学習のfitを説明しよう、と言うことです。言い換えるとモデルがデータをどう扱ったか解釈しよう、と言うことです。なんとなくまとめを先に行っておくと次の通りです。ブラックボックス機械学習も僕の知っている２年前に比べて色々と賑やかになってきました。決定木だとLightGBMやXGBOOSTありますし、ニューラルネットだとStacked LSTMやGANとかですねえ。そんな優秀なアルゴリズム達をどうやって理解すればいいんだろう？と言うのがブラックボックス問題です。例えば「ニューラルネットで学習させたけどこの重みの意味って、、、？」とか「この入力と出力の間の関係は、、？」とか「この変数って予測にプラスに働いたのか、、、？」などで</p> <ul class="entrysearch-entry-tags"> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://research.miidas.jp/2019/11/shap%E5%80%A4%E3%81%A7%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8B%E5%89%8D%E3%81%ABpermutation-importance%E3%82%92%E7%9F%A5%E3%82%8B/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry "> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://data-analysis-stats.jp/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%83%A2%E3%83%87%E3%83%AB%E3%82%92%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8Bshap/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://data-analysis-stats.jp/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%83%A2%E3%83%87%E3%83%AB%E3%82%92%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8Bshap/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fdata-analysis-stats.jp%2F%25E6%25A9%259F%25E6%25A2%25B0%25E5%25AD%25A6%25E7%25BF%2592%2F%25E6%25A9%259F%25E6%25A2%25B0%25E5%25AD%25A6%25E7%25BF%2592%25E3%2583%25A2%25E3%2583%2587%25E3%2583%25AB%25E3%2582%2592%25E8%25A7%25A3%25E9%2587%2588%25E3%2581%2599%25E3%2582%258Bshap%2F" alt="">機械学習モデルを解釈するSHAP – S-Analysis </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/data-analysis-stats.jp/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%83%A2%E3%83%87%E3%83%AB%E3%82%92%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8Bshap/" class="js-keyboard-entry-page-openable" title="機械学習モデルを解釈するSHAP – S-Analysis (4 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 4 users </a> </span> </li> <li> <a href="/site/data-analysis-stats.jp/" title="『data-analysis-stats.jp』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> data-analysis-stats.jp </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2022/11/17</span></li> </ul> </div> <div class="centerarticle-entry-contents "> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">機械学習モデルを学習させた時に、実際にモデルはどの特徴量を見て予測をしているのかが知りたい時があります。今回はモデルによる予測結果の解釈性を向上させる方法の１つであるSHAPを解説します。目次1. XAIとは 2. SHAPとは 3. 実験・コード 1：回帰モデル（Diabetes dataset）＿＿3.1 データ読み込み＿＿3.2 モデル作成＿＿3.3 SHAP値＿＿3.4 SHAP可視化 4. 実験・コード 2：画像データ（Imagenet）＿＿4.1 データ読み込み＿＿4.2 モデル作成＿＿4.3 SHAP可視化 1. XAI (Explainable AI)とはXAI はExplainable AI（説明可能なAI）の英略称です。言葉通り、予測結果や推定結果に至るプロセスが人間によって説明可能になっている機械学習のモデルに関する技術や研究分野のことを指します。　A</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://data-analysis-stats.jp/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92%E3%83%A2%E3%83%87%E3%83%AB%E3%82%92%E8%A7%A3%E9%87%88%E3%81%99%E3%82%8Bshap/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://www.medi-08-data-06.work/entry/how_to_use_shap"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://www.medi-08-data-06.work/entry/how_to_use_shap" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fwww.medi-08-data-06.work%2Fentry%2Fhow_to_use_shap" alt="">SHAPを使って機械学習モデルと対話する - 医療職からデータサイエンティストへ </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/www.medi-08-data-06.work/entry/how_to_use_shap" class="js-keyboard-entry-page-openable" title="SHAPを使って機械学習モデルと対話する - 医療職からデータサイエンティストへ (3 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 3 users </a> </span> </li> <li> <a href="/site/www.medi-08-data-06.work/" title="『www.medi-08-data-06.work』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> www.medi-08-data-06.work </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2019/12/02</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">機械学習モデルは、統計モデルよりも予測に長けた手法であり、皆様もご存知の通り様々な場面で用いられています。一方で、結果の解釈の面ではブラックボックスになりやすいため、モデルの作成時のみならず、機械学習に覚えのない方々とコミュニュケーションをする上でも重要な課題です。そんな、機械学習モデルと対話するためのツールがSAHP値(SHapley Additive exPlanation Values)です。SHAPを使うと、機械学習モデルが特徴量をどのように使って予測をしたのか、特徴量は予測結果にどれぐらい影響を与えているのか、などをデータ全体(Global)、さらに個別のサンプルごと(Individual)に確認することができます。今回はSHAPを使って、学習した機械学習モデルと対話してみましょう。今回は主に以下を参考にしました。 github.com shap.readthedocs.io</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/python" data-gtm-click-label="entry-search-result-item-tag">python</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://www.medi-08-data-06.work/entry/how_to_use_shap"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://www.medi-08-data-06.work/entry/how_to_use_shap" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/af463283950cf7ce7f1301af4414773d10936d82/height=288;version=1;width=512/https%3A%2F%2Fcdn.image.st-hatena.com%2Fimage%2Fscale%2F449f357d7de7aeb6a3a2f07bd7d78963ba078e2c%2Fbackend%3Dimagemagick%3Bheight%3D1200%3Bversion%3D1%3Bwidth%3D1200%2Fhttps%253A%252F%252Fcdn-ak.f.st-hatena.com%252Fimages%252Ffotolife%252Fh%252Fh-wadsworth02%252F20191126%252F20191126130819.jpg" alt="SHAPを使って機械学習モデルと対話する - 医療職からデータサイエンティストへ" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191?s=09"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191?s=09" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fqiita.com%2Fs1ok69oo%2Fitems%2F0bf92b84e565789a2191%3Fs%3D09" alt="">SHAPで因果関係を説明できる？ - Qiita </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/qiita.com/s1ok69oo/items/0bf92b84e565789a2191?s=09" class="js-keyboard-entry-page-openable" title="SHAPで因果関係を説明できる？ - Qiita (3 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 3 users </a> </span> </li> <li> <a href="/site/qiita.com/s1ok69oo" title="『qiita.com/s1ok69oo』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> qiita.com/s1ok69oo </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2024/08/07</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">はじめに予測モデル（機械学習モデル）を解釈するのに有用なSHAPを用いて因果関係を説明することができるか、についてPythonによるシミュレーションを交えてまとめました。内容に誤り等ございましたら、ご指摘いただけますと幸いです。結論基本的に、SHAPで因果関係は説明できません。これは、SHAPが予測モデルの因果ではなく相関を明らかにするものであるからです。そこで今回は、予測モデルをSHAPで解釈する上でありがちなミスリーディングや、それに関連する因果効果を推定するためのアプローチについて記載しています。そもそもSHAPとは SHAPとはSHapley Additive exPlanationsの略で、協力ゲーム理論のShapley Valueを機械学習に応用した手法です。「その予測モデルがなぜ、その予測値を算出しているか」を解釈するためのツールとしてオープンソースのライブラリが開</p> <ul class="entrysearch-entry-tags"> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191?s=09"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://qiita.com/s1ok69oo/items/0bf92b84e565789a2191?s=09" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/afd2b30384de3523c08f78d0dddb66b12555d47e/height=288;version=1;width=512/https%3A%2F%2Fqiita-user-contents.imgix.net%2Fhttps%253A%252F%252Fcdn.qiita.com%252Fassets%252Fpublic%252Farticle-ogp-background-412672c5f0600ab9a64263b751f1bc81.png%3Fixlib%3Drb-4.0.0%26w%3D1200%26mark64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTk3MiZoPTM3OCZ0eHQ9U0hBUCVFMyU4MSVBNyVFNSU5QiVBMCVFNiU5RSU5QyVFOSU5NiVBMiVFNCVCRiU4MiVFMyU4MiU5MiVFOCVBQSVBQyVFNiU5OCU4RSVFMyU4MSVBNyVFMyU4MSU4RCVFMyU4MiU4QiVFRiVCQyU5RiZ0eHQtYWxpZ249bGVmdCUyQ3RvcCZ0eHQtY29sb3I9JTIzMUUyMTIxJnR4dC1mb250PUhpcmFnaW5vJTIwU2FucyUyMFc2JnR4dC1zaXplPTU2JnM9NzExZTZlZTNkZWI5Mzg2Yjk0N2Q5YmM4OTQxMWE1ZGY%26mark-x%3D142%26mark-y%3D57%26blend64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZoPTc2Jnc9NzcwJnR4dD0lNDBzMW9rNjlvbyZ0eHQtY29sb3I9JTIzMUUyMTIxJnR4dC1mb250PUhpcmFnaW5vJTIwU2FucyUyMFc2JnR4dC1zaXplPTM2JnR4dC1hbGlnbj1sZWZ0JTJDdG9wJnM9Y2NkOTI4ZDY0YzBjMzFlY2U0ZWNjNmQxZGRiMGNiNTQ%26blend-x%3D142%26blend-y%3D486%26blend-mode%3Dnormal%26s%3Daafe3a34fafecfbace22fe19baf9ad3a" alt="SHAPで因果関係を説明できる？ - Qiita" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry "> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://datadriven-rnd.com/shap/"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://datadriven-rnd.com/shap/" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fdatadriven-rnd.com%2Fshap%2F" alt="">SHAP を用いた機械学習への解釈性付与 </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/datadriven-rnd.com/shap/" class="js-keyboard-entry-page-openable" title="SHAP を用いた機械学習への解釈性付与 (3 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 3 users </a> </span> </li> <li> <a href="/site/datadriven-rnd.com/" title="『datadriven-rnd.com』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> datadriven-rnd.com </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2023/03/31</span></li> </ul> </div> <div class="centerarticle-entry-contents "> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">（noteの「クリエイターを応援する」で応援していただけると嬉しいです。よろしくお願いします） SHAP（SHapley Additive exPlanations）とは背景昨今では機械学習モデルに解釈性や説明性が強く求められるようになっており「説明可能なAI（Explainable AI｜XAI）」が着目されるようになっている。 SHAPは、予測モデルに対してそのような解釈性を付与するために作られたライブラリ。 SHAPとは協力ゲーム理論のシャープレイ値（Shapley Value）を機械学習に応用したオープンソースのライブラリ。 SHAPを使うことで、ある機械学習モデルが導き出した予測値に対して、それぞれの特徴量の影響をどれだけ受けたか（寄与の度合い）を求めることができる。特に、変数重要度をデータ個別に出せる点が、SHAPの魅力的な点と言える。アウトプット例として、例えば以下の図。</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://datadriven-rnd.com/shap/"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://note.com/dd_techblog/n/n18d354b4defa"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://note.com/dd_techblog/n/n18d354b4defa" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fnote.com%2Fdd_techblog%2Fn%2Fn18d354b4defa" alt="">AIを理解する技術ーSHAPの原理と実装ー｜Dentsu Digital Tech Blog </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/note.com/dd_techblog/n/n18d354b4defa" class="js-keyboard-entry-page-openable" title="AIを理解する技術ーSHAPの原理と実装ー｜Dentsu Digital Tech Blog (3 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 3 users </a> </span> </li> <li> <a href="/site/note.com/dd_techblog" title="『note.com/dd_techblog』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> note.com/dd_techblog </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2022/11/17</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">電通デジタルでデータサイエンティストをしている福田です。これはDentsu Digital Advent Calendar 2021の10日目の記事です。本記事ではXAIの手法の1つであるSHAPについて解説したいと思います。 XAIとはXAIとはexplainable AIの略で、説明可能なAIまたはAIの予測結果を説明する技術のことを指します。モデルの解釈性と精度は基本的にはトレードオフの関係にあり、重回帰や決定木のようなわかりやすいモデルがある一方で、精度を求めようとするとXGBoost、LightGBMのような複雑なモデルに頼ることが多いと思います。XAIは後者のような精度は高いが解釈性が低いモデルをターゲットとしています。 SHAPSHAP(SHapley Additive exPlanation)とは局所的なモデルの説明(1行のデータに対する説明)に該当します。予測値に対し</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> <li><a href="/q/AI" data-gtm-click-label="entry-search-result-item-tag">AI</a></li> <li><a href="/q/technology" data-gtm-click-label="entry-search-result-item-tag">technology</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://note.com/dd_techblog/n/n18d354b4defa"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://note.com/dd_techblog/n/n18d354b4defa" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/c68d776d45273dfd6e4c37e732ef6d6f5b7cb706/height=288;version=1;width=512/https%3A%2F%2Fassets.st-note.com%2Fproduction%2Fuploads%2Fimages%2F67256610%2Frectangle_large_type_2_67afab4d26af17e8b60570eff5e11470.jpg%3Ffit%3Dbounds%26quality%3D85%26width%3D1280" alt="AIを理解する技術ーSHAPの原理と実装ー｜Dentsu Digital Tech Blog" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://qiita.com/FukuharaYohei/items/5f587210e68645809958"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://qiita.com/FukuharaYohei/items/5f587210e68645809958" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fqiita.com%2FFukuharaYohei%2Fitems%2F5f587210e68645809958" alt="">機械学習モデル解釈手法概要(PFI, PD, ICE, SHAP) - Qiita </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/qiita.com/FukuharaYohei/items/5f587210e68645809958" class="js-keyboard-entry-page-openable" title="機械学習モデル解釈手法概要(PFI, PD, ICE, SHAP) - Qiita (3 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 3 users </a> </span> </li> <li> <a href="/site/qiita.com/FukuharaYohei" title="『qiita.com/FukuharaYohei』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> qiita.com/FukuharaYohei </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2022/06/14</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">機械学習のモデル解釈手法を学びました。ほとんどは、書籍「機械学習を解釈する技術」から学んでいます(良書！)。ここに書いているのは、個人的視点でつまみ食いしている感じなので、興味を持ったら購入おすすめです。一部、scikit-learnからも学んでいます。内容全般解釈手法一覧下表が解釈手法の一覧です。解釈手法解釈手法(正式名) 内容視点</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E6%A9%9F%E6%A2%B0%E5%AD%A6%E7%BF%92" data-gtm-click-label="entry-search-result-item-tag">機械学習</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://qiita.com/FukuharaYohei/items/5f587210e68645809958"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://qiita.com/FukuharaYohei/items/5f587210e68645809958" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/6d33f6506ac06b847cb6eece6fa409b4f72d2af3/height=288;version=1;width=512/https%3A%2F%2Fqiita-user-contents.imgix.net%2Fhttps%253A%252F%252Fcdn.qiita.com%252Fassets%252Fpublic%252Farticle-ogp-background-412672c5f0600ab9a64263b751f1bc81.png%3Fixlib%3Drb-4.0.0%26w%3D1200%26mark64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTk3MiZoPTM3OCZ0eHQ9JUU2JUE5JTlGJUU2JUEyJUIwJUU1JUFEJUE2JUU3JUJGJTkyJUUzJTgzJUEyJUUzJTgzJTg3JUUzJTgzJUFCJUU4JUE3JUEzJUU5JTg3JTg4JUU2JTg5JThCJUU2JUIzJTk1JUU2JUE2JTgyJUU4JUE2JTgxJTI4UEZJJTJDJTIwUEQlMkMlMjBJQ0UlMkMlMjBTSEFQJTI5JTIwJnR4dC1jb2xvcj0lMjMyMTIxMjEmdHh0LWZvbnQ9SGlyYWdpbm8lMjBTYW5zJTIwVzYmdHh0LXNpemU9NTYmdHh0LWFsaWduPWxlZnQlMkN0b3Amcz1mYjhmNjRmOTUzMGIwMmNlZDFiMDRmZjlmODRjOGJmZA%26mark-x%3D142%26mark-y%3D57%26blend64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZoPTc2Jnc9NzcwJnR4dD0lNDBGdWt1aGFyYVlvaGVpJnR4dC1jb2xvcj0lMjMyMTIxMjEmdHh0LWZvbnQ9SGlyYWdpbm8lMjBTYW5zJTIwVzYmdHh0LXNpemU9MzYmdHh0LWFsaWduPWxlZnQlMkN0b3Amcz1kYjYyZWY3Y2YwMTczYjVhNDVkOGYzMTJjNDFiM2QwYg%26blend-x%3D142%26blend-y%3D486%26blend-mode%3Dnormal%26s%3Dab4bb1d1ede6f917d81478a431c77812" alt="機械学習モデル解釈手法概要(PFI, PD, ICE, SHAP) - Qiita" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://qiita.com/perico_v1/items/fbbb18681ecc362a4f9e"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://qiita.com/perico_v1/items/fbbb18681ecc362a4f9e" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fqiita.com%2Fperico_v1%2Fitems%2Ffbbb18681ecc362a4f9e" alt="">SHAP(SHapley Additive exPlanation)についての備忘録 - Qiita </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/qiita.com/perico_v1/items/fbbb18681ecc362a4f9e" class="js-keyboard-entry-page-openable" title="SHAP(SHapley Additive exPlanation)についての備忘録 - Qiita (3 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 3 users </a> </span> </li> <li> <a href="/site/qiita.com/perico_v1" title="『qiita.com/perico_v1』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> qiita.com/perico_v1 </a> </li> <li> <a href="/hotentry/it" data-gtm-click-label="entry-search-result-item-category">テクノロジー</a> </li> <li><span class="entry-contents-date">2021/05/24</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">背景・目的ブラックボックス化しがちな機械学習モデルを解釈し、なぜその予測値が出ているのかの説明に役立つSHAP値について、理解を深めるべく論文や公式資料を漁りました。自分用の備忘録としてこちらに内容をまとめておきます。 SHAPとは何か? 正式名称はSHapley Additive exPlanationsで、機械学習モデルの解釈手法の1つなお、「SHAP」は解釈手法自体を指す場合と、手法によって計算された値(SHAP値と呼ぶこともある)を指す場合がある NIPS2017 1にて発表された論文はA Unified Approach to Interpreting Model Predictions それまでに存在した解釈手法(Additive Feature Attribution Methods。 LIMEの基本アイデア)に協力ゲーム理論のShapley Valueを導入して改良し</p> <ul class="entrysearch-entry-tags"> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://qiita.com/perico_v1/items/fbbb18681ecc362a4f9e"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://qiita.com/perico_v1/items/fbbb18681ecc362a4f9e" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/da0175525ef2087a1cc8a075653df46f20735453/height=288;version=1;width=512/https%3A%2F%2Fqiita-user-contents.imgix.net%2Fhttps%253A%252F%252Fcdn.qiita.com%252Fassets%252Fpublic%252Farticle-ogp-background-412672c5f0600ab9a64263b751f1bc81.png%3Fixlib%3Drb-4.0.0%26w%3D1200%26mark64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZ3PTk3MiZoPTM3OCZ0eHQ9U0hBUCUyOFNIYXBsZXklMjBBZGRpdGl2ZSUyMGV4UGxhbmF0aW9uJTI5JUUzJTgxJUFCJUUzJTgxJUE0JUUzJTgxJTg0JUUzJTgxJUE2JUUzJTgxJUFFJUU1JTgyJTk5JUU1JUJGJTk4JUU5JThDJUIyJnR4dC1hbGlnbj1sZWZ0JTJDdG9wJnR4dC1jb2xvcj0lMjMxRTIxMjEmdHh0LWZvbnQ9SGlyYWdpbm8lMjBTYW5zJTIwVzYmdHh0LXNpemU9NTYmcz0zMzJlMzg3NGQyOTkxOTJlOTU4ZTQzNTllNDQxNWFmNQ%26mark-x%3D142%26mark-y%3D57%26blend64%3DaHR0cHM6Ly9xaWl0YS11c2VyLWNvbnRlbnRzLmltZ2l4Lm5ldC9-dGV4dD9peGxpYj1yYi00LjAuMCZoPTc2Jnc9NzcwJnR4dD0lNDBwZXJpY29fdjEmdHh0LWNvbG9yPSUyMzFFMjEyMSZ0eHQtZm9udD1IaXJhZ2lubyUyMFNhbnMlMjBXNiZ0eHQtc2l6ZT0zNiZ0eHQtYWxpZ249bGVmdCUyQ3RvcCZzPWRhMmYxZWNjYjdhNDk3ODAwMTlkZjc3Y2UzYzBkOWE2%26blend-x%3D142%26blend-y%3D486%26blend-mode%3Dnormal%26s%3Dd35058c7144674583f7d0e7c5c951d88" alt="SHAP(SHapley Additive exPlanation)についての備忘録 - Qiita" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> <li class="bookmark-item js-user-bookmark-item js-keyboard-selectable-item"> <div class="centerarticle-entry is-image-entry-unit"> <div class="js-bookmark-stock-button is-hidden"> <button class="js-bookmark-stock-button-ready readlater-button" data-gtm-click-label="entry-search-result-item-stock-button" title="あとで読む" aria-label="あとで読む" tabindex="0" type="button" data-href="https://twitter.com/gapple10086/status/1375024369982787587"><span>あとで読む</span> </button> </div> <div class="centerarticle-entry-header"> <h3 class="centerarticle-entry-title"> <a href="https://twitter.com/gapple10086/status/1375024369982787587" class="js-clickable-link js-keyboard-openable" target="_blank" rel="noopener" data-gtm-click-label="entry-search-result-item-title"> <img class="centerarticle-entry-favicon" src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Ftwitter.com%2Fgapple10086%2Fstatus%2F1375024369982787587" alt="">易水番 on Twitter: "#アズールレーン今日に中国で有名なエロゲーム「アズールレーン」はダウンロードルートを一時閉鎖させられました。原因は中国極端女性権主義が関係部門に告発したからです。具体的な再開時間はまだ公表されていません。 https://t.co/Zx7ShAP9up" </a> </h3> <ul class="centerarticle-entry-data"> <li> <span class="centerarticle-users"> <a href="/entry/s/twitter.com/gapple10086/status/1375024369982787587" class="js-keyboard-entry-page-openable" title="易水番 on Twitter: "#アズールレーン今日に中国で有名なエロゲーム「アズールレーン」はダウンロードルートを一時閉鎖させられました。原因は中国極端女性権主義が関係部門に告発したからです。具体的な再開時間はまだ公表されていません。 https://t.co/Zx7ShAP9up" (3 ブックマーク)" data-gtm-click-label="entry-search-result-item-users"> 3 users </a> </span> </li> <li> <a href="/site/twitter.com/gapple10086" title="『twitter.com/gapple10086』の新着エントリー" data-gtm-click-label="entry-search-result-item-site-search-url"> twitter.com/gapple10086 </a> </li> <li> <a href="/hotentry/economics" data-gtm-click-label="entry-search-result-item-category">政治と経済</a> </li> <li><span class="entry-contents-date">2021/03/26</span></li> </ul> </div> <div class="centerarticle-entry-contents has-image"> <div class="centerarticle-entry-item"> <p class="centerarticle-entry-summary">#アズールレーン今日に中国で有名なエロゲーム「アズールレーン」はダウンロードルートを一時閉鎖させられました。原因は中国極端女性権主義が関係部門に告発したからです。具体的な再開時間はまだ公表されていません。 https://t.co/Zx7ShAP9up</p> <ul class="entrysearch-entry-tags"> <li><a href="/q/%E4%B8%AD%E5%9B%BD" data-gtm-click-label="entry-search-result-item-tag">中国</a></li> </ul> <div class="following-bookmarks-container js-following-bookmarks-container" data-entry-url="https://twitter.com/gapple10086/status/1375024369982787587"> <ul class="following-bookmarks js-following-bookmarks"></ul> <script type="text/x-template" id="template-following-bookmark"> <li> <a href="{{ anchor_path }}" class="following-bookmark-link" data-gtm-label="entry-info-followingBookmarks"> <img src="{{ profile_image_url }}" alt="{{ user_name }}" class="following-bookmark-icon"> </a> <span class="following-bookmark-popup"> <span class="following-bookmark-username">{{ user_name }}</span><span class="following-bookmark-timestamp">{{ created }}</span> <span class="following-bookmark-text"> {{ #comment }}<span class="following-bookmark-comment">{{ comment }}</span>{{ /comment }} <ul class="following-bookmark-tags">{{ #tags }}<li>{{ label }}</li>{{ /tags }}</ul> </span> </span> </li> </script> </div> </div> <div class="centerarticle-entry-image-wrapper"> <a href="https://twitter.com/gapple10086/status/1375024369982787587" class="centerarticle-entry-image" target="_blank" rel="noopener"> <img src="https://cdn-ak-scissors.b.st-hatena.com/image/square/45915eab7b46575ef8de3d604f2e211fc263f955/height=288;version=1;width=512/https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FExUPZBsVgAMG909.jpg" alt="易水番 on Twitter: "#アズールレーン今日に中国で有名なエロゲーム「アズールレーン」はダウンロードルートを一時閉鎖させられました。原因は中国極端女性権主義が関係部門に告発したからです。具体的な再開時間はまだ公表されていません。 https://t.co/Zx7ShAP9up"" data-gtm-click-label="entry-search-result-item-image"> </a> </div> </div> </div> </li> </ul> <div class="centerarticle-pager is-single"> <span>1</span> </div> </div> <ins id="entrysearch-ad-rectangle-middle-sp" class="entrylist-ad-sp-fluid for-sp"></ins> <div id="entrysearch-ad-rectangle-bottom-sp" class="touch-ad-rectangle"></div> </div> <div class="entry-wrapper"> <section class="entry-group"> <h2 class="entry-group-sectionTitle">新着記事</h2> <ul class="entry-group-list"> <li> <div class="entry-group-list-inner"> <div class="entry-group-text"> <h3 class="entry-group-title"><a href="/entry/s/hypergadget.jp/xiaomi14tpro_amazon_sale_end2411/" title="【悲報】Xiaomi 14T Proの『12GB/512GBモデル』がAmazonで割引終了か | ハイパーガジェット通信" data-gtm-label="entry-recentEntriesInEntrySearch-title">【悲報】Xiaomi 14T Proの『12GB/512GBモデル』がAmazonで割引終了か | ハイパーガジェット通信</a></h3> <p class="entry-group-meta"><span class="entry-group-users"><a href="/entry/s/hypergadget.jp/xiaomi14tpro_amazon_sale_end2411/" data-gtm-label="entry-recentEntriesInEntrySearch-users">1 user</a></span><span class="entry-group-domain"><a href="/site/hypergadget.jp/" data-gtm-label="entry-recentEntriesInEntrySearch-domain"><img src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fhypergadget.jp%2Fxiaomi14tpro_amazon_sale_end2411%2F" alt="">hypergadget.jp</a></span></p> </div> </div> </li> <li> <div class="entry-group-list-inner"> <div class="entry-group-text"> <h3 class="entry-group-title"><a href="/entry/s/www.utanutan.com/entry/2024/11/26/173400" title="きっとDNAがそうさせるのよね - 独身オバチャンの猫とのまったり節約生活" data-gtm-label="entry-recentEntriesInEntrySearch-title">きっとDNAがそうさせるのよね - 独身オバチャンの猫とのまったり節約生活</a></h3> <p class="entry-group-meta"><span class="entry-group-users"><a href="/entry/s/www.utanutan.com/entry/2024/11/26/173400" data-gtm-label="entry-recentEntriesInEntrySearch-users">1 user</a></span><span class="entry-group-domain"><a href="/site/www.utanutan.com/" data-gtm-label="entry-recentEntriesInEntrySearch-domain"><img src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fwww.utanutan.com%2Fentry%2F2024%2F11%2F26%2F173400" alt="">www.utanutan.com</a></span></p> </div> </div> </li> <li> <div class="entry-group-list-inner"> <div class="entry-group-text"> <h3 class="entry-group-title"><a href="/entry/s/www.lettuceclub.net/news/article/1224936/" title="「寝ている熊を起こさないで」北欧女子が解説！スウェーデンの面白いことわざ - レタスクラブ" data-gtm-label="entry-recentEntriesInEntrySearch-title">「寝ている熊を起こさないで」北欧女子が解説！スウェーデンの面白いことわざ - レタスクラブ</a></h3> <p class="entry-group-meta"><span class="entry-group-users"><a href="/entry/s/www.lettuceclub.net/news/article/1224936/" data-gtm-label="entry-recentEntriesInEntrySearch-users">1 user</a></span><span class="entry-group-domain"><a href="/site/www.lettuceclub.net/" data-gtm-label="entry-recentEntriesInEntrySearch-domain"><img src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fwww.lettuceclub.net%2Fnews%2Farticle%2F1224936%2F" alt="">www.lettuceclub.net</a></span></p> </div> </div> </li> <li> <div class="entry-group-list-inner"> <div class="entry-group-text"> <h3 class="entry-group-title"><a href="/entry/s/femimatsu.com/article/505800010.html" title="【またかよ】フェミニストさん、透明なフィギュアに発狂" data-gtm-label="entry-recentEntriesInEntrySearch-title">【またかよ】フェミニストさん、透明なフィギュアに発狂</a></h3> <p class="entry-group-meta"><span class="entry-group-users"><a href="/entry/s/femimatsu.com/article/505800010.html" data-gtm-label="entry-recentEntriesInEntrySearch-users">1 user</a></span><span class="entry-group-domain"><a href="/site/femimatsu.com/" data-gtm-label="entry-recentEntriesInEntrySearch-domain"><img src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Ffemimatsu.com%2Farticle%2F505800010.html" alt="">femimatsu.com</a></span></p> </div> </div> </li> <li> <div class="entry-group-list-inner"> <div class="entry-group-text"> <h3 class="entry-group-title"><a href="/entry/s/ascii.jp/elem/000/004/235/4235868/2/" title="業界トップ級の画像生成AIツール「Recraft」　Adobeユーザーは体験する価値あり (2/9)" data-gtm-label="entry-recentEntriesInEntrySearch-title">業界トップ級の画像生成AIツール「Recraft」　Adobeユーザーは体験する価値あり (2/9)</a></h3> <p class="entry-group-meta"><span class="entry-group-users"><a href="/entry/s/ascii.jp/elem/000/004/235/4235868/2/" data-gtm-label="entry-recentEntriesInEntrySearch-users">1 user</a></span><span class="entry-group-domain"><a href="/site/ascii.jp/" data-gtm-label="entry-recentEntriesInEntrySearch-domain"><img src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fascii.jp%2Felem%2F000%2F004%2F235%2F4235868%2F2%2F" alt="">ascii.jp</a></span></p> </div> </div> </li> <li> <div class="entry-group-list-inner"> <div class="entry-group-text"> <h3 class="entry-group-title"><a href="/entry/s/etemity.hatenablog.com/entry/2024/11/26/164211" title="岩に刻まれた地図【ティティウム渓谷】【黒い砂漠】 - 黒い砂漠の永遠　Black Desert" data-gtm-label="entry-recentEntriesInEntrySearch-title">岩に刻まれた地図【ティティウム渓谷】【黒い砂漠】 - 黒い砂漠の永遠　Black Desert</a></h3> <p class="entry-group-meta"><span class="entry-group-users"><a href="/entry/s/etemity.hatenablog.com/entry/2024/11/26/164211" data-gtm-label="entry-recentEntriesInEntrySearch-users">1 user</a></span><span class="entry-group-domain"><a href="/site/etemity.hatenablog.com/" data-gtm-label="entry-recentEntriesInEntrySearch-domain"><img src="https://cdn-ak2.favicon.st-hatena.com/64?url=https%3A%2F%2Fetemity.hatenablog.com%2Fentry%2F2024%2F11%2F26%2F164211" alt="">etemity.hatenablog.com</a></span></p> </div> </div> </li> </ul> </section> </div> </div> <div id="keyboard-shortcut-help" class="shortcut-help is-hidden"> <h3 class="shortcut-help-title">キーボードショートカット一覧</h3> <p><span class="shortcut-help-key">j</span>次のブックマーク</p> <p><span class="shortcut-help-key">k</span>前のブックマーク</p> <p><span class="shortcut-help-key">l</span>あとで読む</p> <p><span class="shortcut-help-key">e</span>コメント一覧を開く</p> <p><span class="shortcut-help-key">o</span>ページを開く</p> <button tyle="button" class="shortcut-help-close js-close-button">✕</button> </div> <script type="text/javascript"> // ---------------- dfp section ---------------- var valve = valve || []; if (isMobile()) { valve.push(function(v) { v.defineDFPSlot({ unit: '/4374287/bookmark_overlay_sp', sizes: [[320,80],[320,70],[320,50]], slotId: 'touch-footer-ad' }); v.defineDFPSlot({ unit: '/4374287/bookmark_entrysearch_rectangle_bottom_sp', sizes: [[300,250]], slotId: 'entrysearch-ad-rectangle-bottom-sp' }); v.sealDFPSlots(); }); } else { valve.push(function(v) { v.defineDFPSlot({ unit: '/4374287/bookmark_entrysearch_sidebar_top_pc', sizes: [[300,250]], slotId: 'entrysearch-ad-sidebar-top' }); v.defineDFPSlot({ unit: '/4374287/bookmark_entrysearch_sidebar_bottom_pc', sizes: [[300,250]], slotId: 'entrysearch-ad-sidebar-bottom' }); v.sealDFPSlots(); }); } // ---------------- ydn section ---------------- if (isMobile()) { valve.push(function(v) { v.defineFluidSlot({ insId: 'entrysearch-ad-rectangle-middle-sp', yadsAdDs: '62199_242886' }); }); } </script> <script> var valve = window.valve || []; valve.push(function(v) { var pageKeyValues = {}; if (document.cookie.split(';').some(function(item) { return item.indexOf('202011_guest_status=register_incomplete') >= 0})) { pageKeyValues.bookmark_status = 'register_incomplete'; } v.config({ pageKeyValues: pageKeyValues, textSelector: ".entry-search-articles", content: { result: "ng" }, service: 'bookmark' }); }); </script> <script src="https://cdn.pool.st-hatena.com/valve/valve.js" async></script> <div class="footer"> <div class="footer-inner"> <div class="footer-inner-group footer-bookmark"> <h2 class="footer-title"><a href="/" data-gtm-label="footer-btop">はてなブックマーク</a></h2> <div class="footer-list-wrapper"> <ul class="footer-list footer-list--pc"> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/all" class="hotentry js-keyboard-openable" data-gtm-label="footerCategories-hotentry">総合</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/general" class="general js-keyboard-openable" data-gtm-label="footerCategories-general">一般</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/social" class="cat-social js-keyboard-openable" data-gtm-label="footerCategories-social">世の中</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/economics" class="cat-economics js-keyboard-openable" data-gtm-label="footerCategories-economics">政治と経済</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/life" class="cat-life js-keyboard-openable" data-gtm-label="footerCategories-life">暮らし</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/knowledge" class="cat-knowledge js-keyboard-openable" data-gtm-label="footerCategories-knowledge">学び</a></li> </ul> <ul class="footer-list footer-list--pc"> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/it" class="cat-it js-keyboard-openable" data-gtm-label="footerCategories-it">テクノロジー</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/entertainment" class="cat-entertainment js-keyboard-openable" data-gtm-label="footerCategories-entertainment">エンタメ</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/game" class="cat-game js-keyboard-openable" data-gtm-label="footerCategories-game">アニメとゲーム</a></li> <li class="footer-list-item js-keyboard-selectable-item"><a href="/hotentry/fun" class="cat-fun js-keyboard-openable" data-gtm-label="footerCategories-fun">おもしろ</a></li> </ul> <ul class="footer-list"> <li class="footer-list-item"><a href="/guide/tools">アプリ・拡張機能</a></li> <li class="footer-list-item"><a href="https://bookmark.hatenastaff.com/">開発ブログ</a></li> <li class="footer-list-item"><a href="/help/">ヘルプ</a></li> <li class="footer-list-item"><a href="https://hatena.zendesk.com/hc/ja/categories/900000153166">お問い合わせ</a></li> <li class="footer-list-item footer-list-item--sp"><a href="/guide/guideline">ガイドライン</a></li> <li class="footer-list-item footer-list-item--sp"><a href="https://policies.hatena.ne.jp/rule">利用規約</a></li> <li class="footer-list-item footer-list-item--sp"><a href="https://policies.hatena.ne.jp/privacypolicy-ja">プライバシーポリシー</a></li> <li class="footer-list-item footer-list-item--sp"><a href="https://policies.hatena.ne.jp/cookie-policy">利用者情報の外部送信について</a></li> </ul> <ul class="footer-list footer-list--pc"> <li class="footer-list-item footer-list-item--pc"><a href="/guide/guideline">ガイドライン</a></li> <li class="footer-list-item footer-list-item--pc"><a href="https://policies.hatena.ne.jp/rule">利用規約</a></li> <li class="footer-list-item footer-list-item--pc"><a href="https://policies.hatena.ne.jp/privacypolicy-ja">プライバシーポリシー</a></li> <li class="footer-list-item footer-list-item--pc"><a href="https://policies.hatena.ne.jp/cookie-policy">利用者情報の外部送信について</a></li> </ul> </div> </div> <div class="footer-inner-group footer-bookmark-sns"> <h2 class="footer-title">公式Twitter</h2> <div class="footer-bookmark-list"> <ul class="footer-list"> <li class="footer-list-item"><a href="https://twitter.com/HatenaBookmark" target="_blank" rel="noopener" data-gtm-click-label="footer-officialTwitter-name">公式アカウント</a></li> <li class="footer-list-item"><a href="https://twitter.com/hatebu" target="_blank" rel="noopener" data-gtm-click-label="footer-hotentryTwitter-name">ホットエントリー</a></li> </ul> </div> </div> <div class="footer-inner-group footer-hatena"> <h2 class="footer-title">はてなのサービス</h2> <ul class="footer-list"> <li class="footer-list-item"><a href="https://hatenablog.com/" data-gtm-label="footerServices-blog">はてなブログ</a></li> <li class="footer-list-item"><a href="https://hatenablog.com/guide/pro?plus_via=hatenabookmark_footer&utm_source=hatenabookmark&utm_medium=referral&utm_campaign=hatenabookmark_to_hatenablog_pro&utm_content=hatenabookmark_footer" data-gtm-label="footerServices-blog-pro">はてなブログPro</a></li> <li class="footer-list-item"><a href="https://q.hatena.ne.jp/" data-gtm-label="footerServices-question">人力検索はてな</a></li> <li class="footer-list-item"><a href="http://d.hatena.ne.jp/" data-gtm-label="footerServices-keyword">はてなブログタグ</a></li> <li class="footer-list-item"><a href="http://hatenanews.com/" data-gtm-label="footerServices-hatenanews">はてなニュース</a></li> <li class="footer-list-item"><a href="https://soredoko.jp/" data-gtm-label="footerServices-soredoko">ソレドコ</a></li> </ul> </div> </div> <div class="footer-bottom"> <div class="footer-app"> <ul> <li class="footer-list-item"><a href="https://apps.apple.com/jp/app/hatenabukkumaku/id354976659?mt=8" target="_blank" rel="noopener" data-gtm-click-label="footer-app-ios"><img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/common/dl-button/appstore@2x.png" alt="App Storeからダウンロード"></a></li> <li class="footer-list-item"><a href="https://play.google.com/store/apps/details?id=com.hatena.android.bookmark&referrer=utm_source%3Dbookmark_sp%26utm_campaign%3Dbookmark_sp_user&hl=ja" target="_blank" rel="noopener" data-gtm-click-label="footer-app-android"><img src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/images/v4/public/common/dl-button/googleplay@2x.png" alt="Google Playで手に入れよう"></a></li> </ul> </div> <div class="copyright">Copyright © 2005-2024 <a href="http://www.hatena.ne.jp/">Hatena</a>. All Rights Reserved.</div> </div> </div> <div id="touch-footer-ad"></div> <div id="touch-footer-ad-placeholder"></div> <div class="modal-overlay"></div> <script type="text/javascript" src="https://b.st-hatena.com/77d2ee6f79ddaf78eb2d198936bae315b6721993/js/v4/bookmark.star.js" async></script> <div class="js-head-notify is-hidden">設定を変更しました<span class="js-head-notify-close-button">x</span></div> </body> </html>