

{"id":38540,"date":"2019-10-31T00:00:26","date_gmt":"2019-10-30T16:00:26","guid":{"rendered":"https:\/\/case.ntu.edu.tw\/blog\/?p=38540"},"modified":"2022-01-04T00:35:20","modified_gmt":"2022-01-03T16:35:20","slug":"%e5%bc%b7%e5%8c%96%e5%ad%b8%e7%bf%92%e6%a8%a1%e5%9e%8b%e7%9a%84%e8%a8%ba%e6%96%b7%e5%b7%a5%e5%85%b7%e2%94%80%e2%94%80bsuite","status":"publish","type":"post","link":"https:\/\/case.ntu.edu.tw\/blog\/?p=38540","title":{"rendered":"\u5f37\u5316\u5b78\u7fd2\u6a21\u578b\u7684\u8a3a\u65b7\u5de5\u5177\u2500\u2500bsuite"},"content":{"rendered":"<div class=\"single-post-media clr\">\n<div class=\"post-thumbnail\"><img fetchpriority=\"high\" decoding=\"async\" src=\"https:\/\/highscope.ch.ntu.edu.tw\/wordpress\/wp-content\/uploads\/2019\/10\/tool.jpg\" alt=\"\u5f37\u5316\u5b78\u7fd2\u6a21\u578b\u7684\u8a3a\u65b7\u5de5\u5177\u2500\u2500bsuite\" width=\"500\" height=\"333\" \/><\/div>\n<\/div>\n<div class=\"entry clr\">\n<div class=\"pf-content\">\n<p><strong>\u7de8\u8b6f\uff0f\u9673\u5101\u7ff0<\/strong><\/p>\n<p><strong>7\u7a2e\u6838\u5fc3\u80fd\u529b<\/strong><\/p>\n<p>bsuite\u5c07\u4e00\u500b\u5f37\u5316\u5b78\u7fd2\u6a21\u578b\u61c9\u5177\u5099\u7684\u80fd\u529b\u5206\u70ba\uff1a\u57fa\u790e\u3001\u6b78\u56e0\u3001\u63a2\u7d22\u3001\u901a\u5247\u5316\u3001\u8a18\u61b6\u3001\u96dc\u8a0a\u3001\u898f\u6a21\u7b497\u7a2e\uff0c\u4ee5\u4e0b\u6211\u5011\u4f9d\u5e8f\u89e3\u91cb\u9019\u5e7e\u7a2e\u80fd\u529b\u7684\u5167\u6db5\uff1a<\/p>\n<ul>\n<li>\u57fa\u790e\u80fd\u529b\uff08Basic\uff09<\/li>\n<\/ul>\n<p>\u6e2c\u8a66\u6a21\u578b\u662f\u5426\u80fd\u89e3\u6c7a\u4e00\u7cfb\u5217\u7c21\u55ae\u7684\u6c7a\u7b56\u554f\u984c\u548c\u6a19\u6e96\u5206\u6790\uff0c\u85c9\u6b64\u78ba\u8a8d\u6a21\u578b\u5df2\u53ef\u638c\u63e1\u5176\u4e2d\u7684\u734e\u52f5\u6a5f\u5236\uff0c\u9019\u662f\u6240\u6709\u5f37\u5316\u5b78\u7fd2\u6a21\u578b\u61c9\u5177\u5099\u7684\u57fa\u672c\u80fd\u529b\u3002<\/p>\n<ul>\n<li>\u6b78\u56e0\u80fd\u529b\uff08Credit Assignment\uff09<\/li>\n<\/ul>\n<p>\u6a21\u578b\u662f\u5426\u6709\u80fd\u529b\u63a8\u65b7\u51fa\u5230\u5e95\u662f\u904e\u7a0b\u4e2d\u7684\u54ea\u4e00\u500b\u884c\u70ba\u6216\u6c7a\u7b56\u5c0e\u81f4\u4e86\u6700\u5f8c\u7684\u7d50\u679c\uff0c\u627e\u51fa\u5176\u4e2d\u7684\u95dc\u9375\u56e0\u7d20\u3002<\/p>\n<ul>\n<li>\u63a2\u7d22\u80fd\u529b\uff08Exploration\uff09<\/li>\n<\/ul>\n<p>\u5f37\u5316\u5b78\u7fd2\u6a21\u578b\u5e38\u6703\u9762\u81e8\u9019\u6a23\u7684\u5169\u96e3\uff1a\u5617\u8a66\u8207\u5e73\u5e38\u4e0d\u4e00\u6a23\u7684\u884c\u70ba\uff0c\u5411\u5916\u63a2\u7d22\uff08exploration\uff09\uff0c\u5c0b\u6c42\u66f4\u597d\u7684\u89e3\u6cd5\uff1b\u6216\u5229\u7528\uff08exploitation\uff09\u5df2\u77e5\u77e5\u8b58\uff0c\u56fa\u5b88\u65e2\u6709\u7684\u884c\u70ba\u6a21\u5f0f\uff0c\u4fdd\u5b88\u884c\u4e8b\u3002\u6a21\u578b\u662f\u5426\u53ef\u5728\u5169\u8005\u9593\u53d6\u5f97\u5e73\u8861\uff0c\u751a\u81f3\u78ba\u4fdd\u81ea\u5df1\u7684\u6bcf\u4e00\u6b21\u63a2\u7d22\u66f4\u52a0\u6709\u6548\u7387\u3002<\/p>\n<ul>\n<li>\u901a\u5247\u5316\u80fd\u529b\uff08Generalization\uff09<\/li>\n<\/ul>\n<p>\u6a21\u578b\u662f\u5426\u80fd\u5c07\u6240\u5b78\u5ef6\u4f38\uff0c\u61c9\u7528\u5728\u5176\u4ed6\u985e\u4f3c\u4f46\u53c8\u6709\u4e9b\u5fae\u5dee\u7570\u7684\u554f\u984c\u4e0a\u3002<\/p>\n<ul>\n<li>\u8a18\u61b6\u80fd\u529b\uff08Memory\uff09<\/li>\n<\/ul>\n<p>\u8003\u9a57\u6a21\u578b\u80fd\u5426\u85c9\u7531\u4e00\u7cfb\u5217\u89c0\u5bdf\uff0c\u6709\u7b56\u7565\u5730\u898f\u5283\u672a\u4f86\u6c7a\u7b56\u8207\u884c\u52d5\uff0c\u800c\u975e\u50c5\u8003\u616e\u7576\u4e0b\u72c0\u614b\u3002<\/p>\n<ul>\n<li>\u96dc\u8a0a\u8655\u7406\u80fd\u529b\uff08Stochasticity\/Noise\uff09<\/li>\n<\/ul>\n<p>\u9019\u500b\u4e16\u754c\u662f\u5177\u6709\u96a8\u6a5f\u6027\u7684\u3002\u540c\u6a23\u60c5\u5883\u4e0b\u63a1\u53d6\u76f8\u540c\u7684\u884c\u70ba\uff0c\u4f9d\u7136\u6709\u53ef\u80fd\u5f97\u5230\u4e0d\u540c\u7684\u7d50\u679c\u3002\u6a21\u578b\u80fd\u5426\u8655\u7406\u8655\u7406\u9019\u6a23\u7684\u554f\u984c\uff0c\u53bb\u9664\u8cc7\u6599\u4e2d\u7684\u96dc\u8a0a\u3001\u9078\u64c7\u66f4\u597d\u7684\u7b56\u7565\uff0c\u4e5f\u662f\u61c9\u5177\u5099\u7684\u80fd\u529b\u3002<\/p>\n<ul>\n<li>\u5404\u7a2e\u898f\u6a21\u7684\u8655\u7406\u80fd\u529b\uff08Scale\uff09<\/li>\n<\/ul>\n<p>\u9019\u88e1\u7684\u898f\u6a21\u6307\u7684\u662f\u734e\u52f5\uff08reward\uff09\u7684\u5927\u5c0f\u3002\u91dd\u5c0d\u4e0d\u540c\u5927\u5c0f\u7684\u734e\u52f5\uff0c\u6a21\u578b\u61c9\u6709\u9069\u7576\u7684\u8a08\u7b97\u65b9\u6cd5\u3002<\/p>\n<p><strong>\u7c21\u4fbf\u7684\u8a3a\u65b7\u5de5\u5177<\/strong><\/p>\n<p>\u4e00\u500b\u5177\u6709\u4ee3\u8868\u6027\u7684\u5be6\u9a57\u74b0\u5883\u9808\u540c\u6642\u7b26\u5408\u300c\u76ee\u6a19\u660e\u78ba\u300d\uff08targeted\uff09\u3001\u300c\u7c21\u6f54\u300d\uff08simple\uff09\u3001\u300c\u6709\u6311\u6230\u6027\u300d\uff08challenging\uff09\u3001\u300c\u53ef\u64f4\u5c55\u300d\uff08scalable\uff09\u3001\u300c\u5feb\u901f\u300d\uff08fast\uff09\u7b49\u4e94\u9805\u689d\u4ef6\u3002\u6bcf\u500b\u5be6\u9a57\u74b0\u5883\u50c5\u6aa2\u6e2c7\u500b\u6838\u5fc3\u80fd\u529b\u4e2d\u7684\u4e00\u90e8\u5206\uff0c\u7d9c\u5408\u591a\u500b\u5177\u6709\u4ee3\u8868\u6027\u7684\u5be6\u9a57\u74b0\u5883\uff0c\u4fbf\u53ef\u8a55\u4f30\u6a21\u578b\u7684\u6574\u9ad4\u8868\u73fe\uff0c\u4e26\u4ee5\u96f7\u9054\u5716\u7684\u65b9\u5f0f\u5448\u73fe\u3002<\/p>\n<div id=\"attachment_83170\" class=\"wp-caption aligncenter\"><img decoding=\"async\" class=\"size-full wp-image-83170\" src=\"https:\/\/highscope.ch.ntu.edu.tw\/wordpress\/wp-content\/uploads\/2019\/10\/Cartpole.png\" alt=\"\" width=\"1274\" height=\"330\" \/><\/p>\n<p class=\"wp-caption-text\">\u300cCartPole\u300d\u662fbsuite\u4e2d\u7684\u4e00\u500b\u5be6\u9a57\u5834\u666f\uff0c\u85c9\u7531\u79fb\u52d5\u8eca\u5b50\u4ee5\u5e73\u8861\u8eca\u4e0a\u687f\u67f1\u3002\u8a72\u5834\u666f\u4e3b\u8981\u6aa2\u6e2c\u6a21\u578b\u7684\u57fa\u790e\u80fd\u529b\u3001\u6b78\u56e0\u80fd\u529b\u8207\u901a\u5247\u5316\u80fd\u529b\uff08\u5716\u7247\u4f86\u6e90\uff1aOsband et al., 2019.\uff09<\/p>\n<\/div>\n<p>\u4ee5\u4e0b\u5716\u70ba\u4f8b\uff0c\u5176\u4e2d\u91dd\u5c0d7\u7a2e\u80fd\u529b\u7684\u8a55\u5206\uff0c\u662f\u904b\u884c\u591a\u500b\u5177\u4ee3\u8868\u6027\u7684\u5be6\u9a57\u7d2f\u7a4d\u800c\u6210\uff0c\u7d93\u8996\u89ba\u5316\u5f8c\u6211\u5011\u53ef\u4ee5\u5f88\u5bb9\u6613\u5730\u770b\u51fa\u4e0d\u540c\u985e\u578b\u7684\u6df1\u5ea6\u5f37\u5316\u5b78\u7fd2\u6a21\u578b\u5728\u5404\u9805\u80fd\u529b\u4e0a\u7684\u512a\u52e2\u8207\u5f31\u9ede\u3002\u5c0d\u65bc\u6f14\u7b97\u6cd5\u958b\u767c\u8207\u7814\u7a76\u5718\u968a\u800c\u8a00\uff0c\u80fd\u4ee5\u66f4\u5c11\u7684\u8cc7\u6e90\u8207\u6642\u9593\uff0c\u66f4\u7cbe\u78ba\u5730\u5224\u65b7\u6a21\u578b\u7684\u74f6\u9838\u8207\u7f3a\u9677\uff0cbsuite\u7121\u7591\u662f\u4e00\u5927\u798f\u97f3\u3002<\/p>\n<div id=\"attachment_83171\" class=\"wp-caption aligncenter\"><img decoding=\"async\" class=\"size-full wp-image-83171\" src=\"https:\/\/highscope.ch.ntu.edu.tw\/wordpress\/wp-content\/uploads\/2019\/10\/radar-graph.png\" alt=\"\" width=\"1167\" height=\"744\" \/><\/p>\n<p class=\"wp-caption-text\">\u900f\u904e\u96f7\u9054\u5716\uff0c\u5404\u985e\u6a21\u578b\u7684\u6574\u9ad4\u8868\u73fe\u4e00\u76ee\u4e86\u7136\u3002\u958b\u767c\u6216\u7814\u7a76\u54e1\u4e5f\u53ef\u5229\u7528besuite\u91dd\u5c0d\u7279\u5b9a\u80fd\u529b\u7684\u5206\u6790\u6539\u5584\u6a21\u578b\u52a3\u52e2\u3002\u5716\u4e2d\u7d05\u8272\u4ee3\u8868\u4e86\u96a8\u6a5f\u884c\u70ba\uff0c\u5404\u9805\u5206\u6578\u90fd\u5f88\u4f4e\uff1b\u7da0\u8272\u70baDQN\uff08Deep Q Network\uff09\uff1b\u6dfa\u85cd\u8272\u7dda\u4ee3\u8868\u81ea\u52a9\u5f0fDQN\uff0c\u5176\u5177\u6709\u826f\u597d\u7684\u63a2\u7d22\u80fd\u529b\uff1b\u6df1\u85cd\u8272\u70baActor Critic\uff0c\u5728\u300c\u8a18\u61b6\u300d\u9019\u9805\u80fd\u529b\u4e0a\u5c55\u73fe\u512a\u52e2\u3002(\u5716\u7247\u4f86\u6e90\uff1aOsband et al., 2019.\uff09<\/p>\n<\/div>\n<p>\u60f3\u8981\u8a66\u8a66\u770bbsuite\u9084\u80fd\u505a\u4e9b\u4ec0\u9ebc\u55ce\uff1f\u4f60\u53ef\u4ee5\u5728<a href=\"https:\/\/github.com\/deepmind\/bsuite\">Github\u4e0a<\/a>\u4e0b\u8f09\uff0c\u4e26\u627e\u5230\u76f8\u95dc\u6559\u5b78\u3002\u6559\u5b78\u662f\u4ee5Colab\u5beb\u6210\uff0c\u53ef\u4ee5\u76f4\u63a5\u5728\u700f\u89bd\u5668\u9ad4\u9a57bsuite\u7684\u53b2\u5bb3\u4e4b\u8655\uff0c\u5feb\u4f86\u6aa2\u6e2c\u770b\u770b\u81ea\u5df1\u7684\u6a21\u578b\u5728\u5f37\u5316\u5b78\u7fd2\u5404\u500b\u6838\u5fc3\u80fd\u529b\u7684\u5f97\u5206\u5427\uff01<\/p>\n<p>&nbsp;<\/p>\n<p><strong>\u7de8\u8b6f\u4f86\u6e90<\/strong><\/p>\n<p>Osband et al., \u201c<a href=\"https:\/\/arxiv.org\/pdf\/1908.03568v1.pdf\">Behaviour Suite for Reinforcement Learning<\/a>\u201c, a<em>rXiv.org<\/em>, 2019.<\/p>\n<p><strong>\u53c3\u8003\u8cc7\u6599<\/strong><\/p>\n<p>I. Osband, C. Blundell, A. Pritzel and B. Roy, \u201c<a href=\"http:\/\/papers.nips.cc\/paper\/6501-deep-exploration-via-bootstrapped-dqn.pdf\">Deep Exploration via Bootstrapped DQN<\/a>\u201c,\u00a0<em>Papers.nips.cc<\/em>, 2016.<\/p>\n<p>(\u672c\u6587\u7531\u6559\u80b2\u90e8\u88dc\u52a9\u300cAI\u5831\u5831\u2500AI\u79d1\u666e\u63a8\u5ee3\u8a08\u756b\u300d\u57f7\u884c\u5718\u968a\u7de8\u8b6f)<\/p>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>DeepMind\u8fd1\u671f\u5728Github\u4e00\u53e3\u6c23\u958b\u6e90\u4e86\u4e09\u500b\u8207\u5f37\u5316\u5b78\u7fd2\u76f8\u95dc\u7684\u6846\u67b6\uff0c\u5206\u5225\u662fbsuite\u3001OpenSpiel\u548cSpriteWorld\uff0c\u4e26\u6709\u610f\u5c07\u5176\u4e2d\u7684bsuite\u63a8\u5ee3\u70ba\u5f37\u5316\u5b78\u7fd2\u7684\u6a19\u6e96\u6e2c\u8a66\u6846\u67b6\u3002DeepMind\u6536\u96c6\u4e26\u63d0\u4f9b\u8a31\u591a\u5177\u6709\u4ee3\u8868\u6027\u7684\u5be6\u9a57\u74b0\u5883\uff0c\u4f7f\u5f97bsuite\u6709\u80fd\u529b\u5206\u6790\u6f14\u7b97\u6cd5\u7684\u512a\u52e2\u8207\u52a3\u52e2\uff0c\u4e26\u4ee5\u96f7\u9054\u5716\u7684\u65b9\u5f0f\u5448\u73fe\u3002<\/p>\n","protected":false},"author":21,"featured_media":38541,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3772,3293],"tags":[3456,4989,5806,5807],"aioseo_notices":[],"jetpack_featured_media_url":"https:\/\/case.ntu.edu.tw\/blog\/wp-content\/uploads\/2022\/01\/tool.jpg","_links":{"self":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts\/38540"}],"collection":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/users\/21"}],"replies":[{"embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=38540"}],"version-history":[{"count":1,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts\/38540\/revisions"}],"predecessor-version":[{"id":38542,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/posts\/38540\/revisions\/38542"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=\/wp\/v2\/media\/38541"}],"wp:attachment":[{"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38540"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38540"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/case.ntu.edu.tw\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38540"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}