{"id":4242,"date":"2019-01-10T22:06:54","date_gmt":"2019-01-10T14:06:54","guid":{"rendered":"https:\/\/ixyzero.com\/blog\/?p=4242"},"modified":"2019-01-10T22:06:54","modified_gmt":"2019-01-10T14:06:54","slug":"%e6%9c%ba%e5%99%a8%e5%ad%a6%e4%b9%a0%e7%ae%97%e6%b3%95%e4%b8%adgbdt%e4%b8%8eadaboost%e7%9a%84%e5%8c%ba%e5%88%ab%e4%b8%8e%e8%81%94%e7%b3%bb","status":"publish","type":"post","link":"https:\/\/ixyzero.com\/blog\/archives\/4242.html","title":{"rendered":"\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u4e2dGBDT\u4e0eAdaBoost\u7684\u533a\u522b\u4e0e\u8054\u7cfb"},"content":{"rendered":"<p>=Start=<\/p>\n<h4 id=\"id-\u6a21\u677f-\u7f18\u7531\uff1a\">\u7f18\u7531\uff1a<\/h4>\n<p>\u6700\u8fd1\u5728\u770b\u300a\u673a\u5668\u5b66\u4e60\u300b\u8fd9\u672c\u4e66\uff0c\u60f3\u7b80\u5355\u4e86\u89e3\u4e00\u4e0b\u673a\u5668\u5b66\u4e60\u539f\u7406\u3001\u7b97\u6cd5\u76f8\u5173\u7684\u5185\u5bb9\uff0c\u907f\u514d\u88ab\u65f6\u4ee3\u629b\u4e0b\u3002\u3002\u3002<\/p>\n<p>\u4f46\u65f6\u95f4\u6ca1\u90a3\u4e48\u5145\u88d5\uff0c\u610f\u5fd7\u4e5f\u6ca1\u90a3\u4e48\u575a\u5b9a\uff0c\u770b\u7684\u65ad\u65ad\u7eed\u7eed\u7684\u3002\u4e0d\u8fc7\u8d81\u7740\u4eca\u5929\u6709\u4e9b\u65f6\u95f4\uff0c\u5b66\u4e60\u3001\u6574\u7406\u4e00\u4e0b\u4e4b\u524d\u542c\u7684\u6bd4\u8f83\u591a\u7684GBDT\u3001XGBoost\u3001AdaBoost\u7b49\u7b97\u6cd5\u7684\u6982\u5ff5\uff08\u8fd9\u91cc\u4e0d\u6d89\u53ca\u5177\u4f53\u7684\u516c\u5f0f\u548c\u8bc1\u660e\uff0c\u4e00\u4e2a\u662f\u81ea\u5df1\u73b0\u5728\u4e5f\u770b\u4e0d\u660e\u767d\uff0c\u518d\u4e00\u4e2a\u5c31\u662f\u7f51\u4e0a\u7684\u5f88\u591a\u5176\u4ed6\u6587\u7ae0\u5df2\u7ecf\u8bb2\u7684\u5f88\u591a\u4e86\uff09\uff0c\u4ece\u5174\u8da3\u5165\u624b\uff0c\u4e00\u70b9\u4e00\u70b9\u7684\u79ef\u7d2f\u53ef\u80fd\u66f4\u9002\u5408\u6211\u73b0\u5728\u7684\u72b6\u6001\u3002<\/p>\n<h4 id=\"id-\u6a21\u677f-\u6b63\u6587\uff1a\">\u6b63\u6587\uff1a<\/h4>\n<h5 id=\"id-\u6a21\u677f-\u53c2\u8003\u89e3\u7b54\uff1a\">\u53c2\u8003\u89e3\u7b54\uff1a<\/h5>\n<p>\u5728\u6b63\u5f0f\u8c08\u673a\u5668\u5b66\u4e60\u7b97\u6cd5AdaBoost\u3001GBDT\u4e0eXGBoost\u7684\u533a\u522b\u4e4b\u524d\uff0c\u6709\u5fc5\u8981\u5148\u4e86\u89e3\u4e00\u4e0b<span style=\"color: #ff0000;\"><strong>\u96c6\u6210\u5b66\u4e60\uff08ensemble learning\uff09<\/strong><\/span>\u3002<\/p>\n<p>\u6240\u8c13\u96c6\u6210\u5b66\u4e60\uff0c\u662f\u6307\u901a\u8fc7\u6784\u5efa\u591a\u4e2a\u5206\u7c7b\u5668\uff08\u5f31\u5206\u7c7b\u5668\uff09\u5bf9\u6570\u636e\u96c6\u8fdb\u884c\u9884\u6d4b\uff0c\u7136\u540e\u7528\u67d0\u79cd\u7b56\u7565\u5c06\u591a\u4e2a\u5206\u7c7b\u5668\u9884\u6d4b\u7684\u7ed3\u679c\u96c6\u6210\u8d77\u6765\uff0c\u4f5c\u4e3a\u6700\u7ec8\u9884\u6d4b\u7ed3\u679c\u3002\u901a\u4fd7\u6bd4\u55bb\u5c31\u662f\u201c\u4e09\u4e2a\u81ed\u76ae\u5320\u8d5b\u8fc7\u8bf8\u845b\u4eae\u201d\uff0c\u6216\u4e00\u4e2a\u516c\u53f8\u8463\u4e8b\u4f1a\u4e0a\u7684\u5404\u8463\u4e8b\u6295\u7968\u51b3\u7b56\uff0c\u5b83\u8981\u6c42\u6bcf\u4e2a\u5f31\u5206\u7c7b\u5668\u5177\u5907\u4e00\u5b9a\u7684\u201c\u51c6\u786e\u6027\u201d\uff0c\u5206\u7c7b\u5668\u4e4b\u95f4\u5177\u5907\u201c\u5dee\u5f02\u6027\u201d\u3002\u6709\u65f6\u96c6\u6210\u5b66\u4e60\u4e5f\u88ab\u79f0\u4e3a\u591a\u5206\u7c7b\u5668\u7cfb\u7edf\uff08multi-classifier system\uff09\u3001\u57fa\u4e8e\u59d4\u5458\u4f1a\u7684\u5b66\u4e60\uff08committee-based learning\uff09\u7b49\u3002<\/p>\n<p><strong>\u96c6\u6210\u5b66\u4e60\u6839\u636e\u5404\u4e2a\u5f31\u5206\u7c7b\u5668\u4e4b\u95f4\u6709\u65e0\u4f9d\u8d56\u5173\u7cfb\uff0c\u5206\u4e3aBoosting\u548cBagging\u4e24\u5927\u6d41\u6d3e\uff1a<\/strong><\/p>\n<ul>\n<li><span style=\"color: #ff0000;\"><strong>Boosting\u6d41\u6d3e\uff0c\u5404\u5206\u7c7b\u5668\u4e4b\u95f4\u6709\u4f9d\u8d56\u5173\u7cfb\uff0c\u5fc5\u987b\u4e32\u884c<\/strong><\/span>\uff0c\u6bd4\u5982AdaBoost\u3001GBDT(Gradient Boosting Decision Tree)\u3001XGBoost\u3002<\/li>\n<li><span style=\"color: #ff0000;\"><strong>Bagging\u6d41\u6d3e\uff0c\u5404\u5206\u7c7b\u5668\u4e4b\u95f4\u6ca1\u6709\u4f9d\u8d56\u5173\u7cfb\uff0c\u53ef\u5404\u81ea\u5e76\u884c<\/strong><\/span>\uff0c\u6bd4\u5982\u968f\u673a\u68ee\u6797\uff08Random Forest\uff09\u3002<\/li>\n<\/ul>\n<hr \/>\n<h6>Boosting\u65b9\u6cd5\u7684\u601d\u60f3<\/h6>\n<p><span style=\"color: #ff0000;\">Boosting\u7684\u601d\u60f3\u5176\u5b9e\u76f8\u5f53\u7684\u7b80\u5355\uff0c\u5927\u6982\u662f\uff0c\u5bf9\u4e00\u4efd\u6570\u636e\uff0c\u5efa\u7acbM\u4e2a\u6a21\u578b\uff08\u6bd4\u5982\u5206\u7c7b\uff09\uff0c\u4e00\u822c\u8fd9\u79cd\u6a21\u578b\u6bd4\u8f83\u7b80\u5355\uff0c\u79f0\u4e3a\u5f31\u5206\u7c7b\u5668(weak learner)\uff0c<strong>\u6bcf\u6b21\u5206\u7c7b\u90fd\u5c06\u4e0a\u4e00\u6b21\u5206\u9519\u7684\u6570\u636e\u6743\u91cd\u63d0\u9ad8\u4e00\u70b9\u518d\u8fdb\u884c\u5206\u7c7b<\/strong>\uff0c\u8fd9\u6837\u6700\u7ec8\u5f97\u5230\u7684\u5206\u7c7b\u5668\u5728\u6d4b\u8bd5\u6570\u636e\u4e0e\u8bad\u7ec3\u6570\u636e\u4e0a\u90fd\u53ef\u4ee5\u5f97\u5230\u6bd4\u8f83\u597d\u7684\u6210\u7ee9\u3002<\/span>\u53ef\u4ee5\u60f3\u8c61\u5f97\u5230\uff0c\u7a0b\u5e8f\u8d8a\u5f80\u540e\u6267\u884c\uff0c\u8bad\u7ec3\u51fa\u7684\u6a21\u578b\u5c31\u8d8a\u4f1a\u5728\u610f\u90a3\u4e9b\u5bb9\u6613\u5206\u9519\uff08\u6743\u91cd\u9ad8\uff09\u7684\u70b9\u3002\u5f53\u5168\u90e8\u7684\u7a0b\u5e8f\u6267\u884c\u5b8c\u540e\uff0c\u4f1a\u5f97\u5230M\u4e2a\u6a21\u578b\uff0c\u5206\u522b\u5bf9\u5e94y1(x)&#8230;yM(x)\uff0c\u901a\u8fc7\u52a0\u6743\u7684\u65b9\u5f0f\u7ec4\u5408\u6210\u4e00\u4e2a\u6700\u7ec8\u7684\u6a21\u578bYM(x)\u3002<\/p>\n<p>\u6211\u89c9\u5f97Boosting\u66f4\u50cf\u662f\u4e00\u4e2a\u4eba\u5b66\u4e60\u7684\u8fc7\u7a0b\uff0c\u5f00\u59cb\u5b66\u4e00\u6837\u4e1c\u897f\u7684\u65f6\u5019\uff0c\u4f1a\u53bb\u505a\u4e00\u4e9b\u4e60\u9898\uff0c\u4f46\u662f\u5e38\u5e38\u8fde\u4e00\u4e9b\u7b80\u5355\u7684\u9898\u76ee\u90fd\u4f1a\u5f04\u9519\uff0c\u4f46\u662f\u8d8a\u5230\u540e\u9762\uff0c\u7b80\u5355\u7684\u9898\u76ee\u5df2\u7ecf\u96be\u4e0d\u5012\u4ed6\u4e86\uff0c\u5c31\u4f1a\u53bb\u505a\u66f4\u590d\u6742\u7684\u9898\u76ee\uff0c\u7b49\u5230\u4ed6\u505a\u4e86\u5f88\u591a\u7684\u9898\u76ee\u540e\uff0c\u4e0d\u7ba1\u662f\u96be\u9898\u8fd8\u662f\u7b80\u5355\u7684\u9898\u90fd\u53ef\u4ee5\u89e3\u51b3\u6389\u4e86\u3002<\/p>\n<h6>Gradient Boosting\u65b9\u6cd5<\/h6>\n<p><span style=\"color: #ff0000;\">Gradient Boosting\u662f\u4e00\u79cdBoosting\u7684\u65b9\u6cd5\uff0c\u5b83\u4e3b\u8981\u7684\u601d\u60f3\u662f\uff0c\u6bcf\u4e00\u6b21\u5efa\u7acb\u6a21\u578b\u662f\u5728\u4e4b\u524d\u5efa\u7acb\u6a21\u578b\u635f\u5931\u51fd\u6570\u7684\u68af\u5ea6\u4e0b\u964d\u65b9\u5411\u3002<\/span>\u8fd9\u53e5\u8bdd\u6709\u4e00\u70b9\u62d7\u53e3\uff0c\u635f\u5931\u51fd\u6570(loss function)\u63cf\u8ff0\u7684\u662f\u6a21\u578b\u7684\u4e0d\u9760\u8c31\u7a0b\u5ea6\uff0c\u635f\u5931\u51fd\u6570\u8d8a\u5927\uff0c\u5219\u8bf4\u660e\u6a21\u578b\u8d8a\u5bb9\u6613\u51fa\u9519\uff08\u5176\u5b9e\u8fd9\u91cc\u6709\u4e00\u4e2a\u65b9\u5dee\u3001\u504f\u5dee\u5747\u8861\u7684\u95ee\u9898\uff0c\u4f46\u662f\u8fd9\u91cc\u5c31\u5047\u8bbe\u635f\u5931\u51fd\u6570\u8d8a\u5927\uff0c\u6a21\u578b\u8d8a\u5bb9\u6613\u51fa\u9519\uff09\u3002\u5982\u679c\u6211\u4eec\u7684\u6a21\u578b\u80fd\u591f\u8ba9\u635f\u5931\u51fd\u6570\u6301\u7eed\u7684\u4e0b\u964d\uff0c\u5219\u8bf4\u660e\u6211\u4eec\u7684\u6a21\u578b\u5728\u4e0d\u505c\u7684\u6539\u8fdb\uff0c\u800c\u6700\u597d\u7684\u65b9\u5f0f\u5c31\u662f\u8ba9\u635f\u5931\u51fd\u6570\u5728\u5176\u68af\u5ea6(Gradient)\u7684\u65b9\u5411\u4e0a\u4e0b\u964d\u3002<\/p>\n<hr \/>\n<h5>AdaBoost<\/h5>\n<p>AdaBoost\uff0c\u662f\u82f1\u6587&#8221;Adaptive Boosting&#8221;\uff08\u81ea\u9002\u5e94\u589e\u5f3a\uff09\u7684\u7f29\u5199\uff0c\u7531Yoav Freund\u548cRobert Schapire\u57281995\u5e74\u63d0\u51fa\u3002<strong><span style=\"color: #ff0000;\">\u5b83\u7684\u81ea\u9002\u5e94\u5728\u4e8e\uff1a\u524d\u4e00\u4e2a\u57fa\u672c\u5206\u7c7b\u5668\u5206\u9519\u7684\u6837\u672c\u4f1a\u5f97\u5230\u52a0\u5f3a\uff0c\u52a0\u6743\u540e\u7684\u5168\u4f53\u6837\u672c\u518d\u6b21\u88ab\u7528\u6765\u8bad\u7ec3\u4e0b\u4e00\u4e2a\u57fa\u672c\u5206\u7c7b\u5668\u3002<\/span>\u540c\u65f6\uff0c\u5728\u6bcf\u4e00\u8f6e\u4e2d\u52a0\u5165\u4e00\u4e2a\u65b0\u7684\u5f31\u5206\u7c7b\u5668\uff0c\u76f4\u5230\u8fbe\u5230\u67d0\u4e2a\u9884\u5b9a\u7684\u8db3\u591f\u5c0f\u7684\u9519\u8bef\u7387\u6216\u8fbe\u5230\u9884\u5148\u6307\u5b9a\u7684\u6700\u5927\u8fed\u4ee3\u6b21\u6570\u3002<\/strong><\/p>\n<p>\u5177\u4f53\u8bf4\u6765\uff0c\u6574\u4e2aAdaboost \u8fed\u4ee3\u7b97\u6cd5\u5c313\u6b65\uff1a<\/p>\n<ol>\n<li>\u521d\u59cb\u5316\u8bad\u7ec3\u6570\u636e\u7684\u6743\u503c\u5206\u5e03\u3002\u5982\u679c\u6709N\u4e2a\u6837\u672c\uff0c\u5219\u6bcf\u4e00\u4e2a\u8bad\u7ec3\u6837\u672c\u6700\u5f00\u59cb\u65f6\u90fd\u88ab\u8d4b\u4e88\u76f8\u540c\u7684\u6743\u503c\uff1a1\/N\u3002<\/li>\n<li>\u8bad\u7ec3\u5f31\u5206\u7c7b\u5668\u3002\u5177\u4f53\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u5982\u679c\u67d0\u4e2a\u6837\u672c\u70b9\u5df2\u7ecf\u88ab\u51c6\u786e\u5730\u5206\u7c7b\uff0c\u90a3\u4e48\u5728\u6784\u9020\u4e0b\u4e00\u4e2a\u8bad\u7ec3\u96c6\u4e2d\uff0c\u5b83\u7684\u6743\u503c\u5c31\u88ab\u964d\u4f4e\uff1b\u76f8\u53cd\uff0c\u5982\u679c\u67d0\u4e2a\u6837\u672c\u70b9\u6ca1\u6709\u88ab\u51c6\u786e\u5730\u5206\u7c7b\uff0c\u90a3\u4e48\u5b83\u7684\u6743\u503c\u5c31\u5f97\u5230\u63d0\u9ad8\u3002\u7136\u540e\uff0c\u6743\u503c\u66f4\u65b0\u8fc7\u7684\u6837\u672c\u96c6\u88ab\u7528\u4e8e\u8bad\u7ec3\u4e0b\u4e00\u4e2a\u5206\u7c7b\u5668\uff0c\u6574\u4e2a\u8bad\u7ec3\u8fc7\u7a0b\u5982\u6b64\u8fed\u4ee3\u5730\u8fdb\u884c\u4e0b\u53bb\u3002<\/li>\n<li>\u5c06\u5404\u4e2a\u8bad\u7ec3\u5f97\u5230\u7684\u5f31\u5206\u7c7b\u5668\u7ec4\u5408\u6210\u5f3a\u5206\u7c7b\u5668\u3002\u5404\u4e2a\u5f31\u5206\u7c7b\u5668\u7684\u8bad\u7ec3\u8fc7\u7a0b\u7ed3\u675f\u540e\uff0c\u52a0\u5927\u5206\u7c7b\u8bef\u5dee\u7387\u5c0f\u7684\u5f31\u5206\u7c7b\u5668\u7684\u6743\u91cd\uff0c\u4f7f\u5176\u5728\u6700\u7ec8\u7684\u5206\u7c7b\u51fd\u6570\u4e2d\u8d77\u7740\u8f83\u5927\u7684\u51b3\u5b9a\u4f5c\u7528\uff0c\u800c\u964d\u4f4e\u5206\u7c7b\u8bef\u5dee\u7387\u5927\u7684\u5f31\u5206\u7c7b\u5668\u7684\u6743\u91cd\uff0c\u4f7f\u5176\u5728\u6700\u7ec8\u7684\u5206\u7c7b\u51fd\u6570\u4e2d\u8d77\u7740\u8f83\u5c0f\u7684\u51b3\u5b9a\u4f5c\u7528\u3002\u6362\u8a00\u4e4b\uff0c\u8bef\u5dee\u7387\u4f4e\u7684\u5f31\u5206\u7c7b\u5668\u5728\u6700\u7ec8\u5206\u7c7b\u5668\u4e2d\u5360\u7684\u6743\u91cd\u8f83\u5927\uff0c\u5426\u5219\u8f83\u5c0f\u3002<\/li>\n<\/ol>\n<h5>GBDT(Gradient Boost Decision Tree, \u68af\u5ea6\u589e\u5f3a\u51b3\u7b56\u6811)<\/h5>\n<p>\u53e6\u4e00\u79cdBoosting\u65b9\u6cd5GBDT(Gradient Boost Decision Tree)\uff0c\u5219\u4e0eAdaBoost\u4e0d\u540c\uff0c<b>GBDT\u6bcf\u4e00\u6b21\u7684\u8ba1\u7b97\u662f\u90fd\u4e3a\u4e86\u51cf\u5c11\u4e0a\u4e00\u6b21\u7684\u6b8b\u5dee\uff0c\u8fdb\u800c\u5728\u6b8b\u5dee\u51cf\u5c11\uff08\u8d1f\u68af\u5ea6\uff09\u7684\u65b9\u5411\u4e0a\u5efa\u7acb\u4e00\u4e2a\u65b0\u7684\u6a21\u578b\u3002<\/b><\/p>\n<h5>XGBoost<\/h5>\n<p><b>\u4e8b\u5b9e\u4e0a\uff0c\u5982\u679c\u4e0d\u8003\u8651\u5de5\u7a0b\u5b9e\u73b0\u3001\u89e3\u51b3\u95ee\u9898\u4e0a\u7684\u4e00\u4e9b\u5dee\u5f02\uff0cXGBoost\u4e0eGBDT\u6bd4\u8f83\u5927\u7684\u4e0d\u540c\u5c31\u662f\u76ee\u6807\u51fd\u6570\u7684\u5b9a\u4e49\u3002<\/b>\u5b83\u4eec\u7684\u7b56\u7565\u7c7b\u4f3c\uff0c\u533a\u522b\u5728\u4e8eGBDT\u65e8\u5728\u901a\u8fc7\u4e0d\u65ad\u52a0\u5165\u65b0\u7684\u6811\u6700\u5feb\u901f\u5ea6\u964d\u4f4e\u6b8b\u5dee\uff0c\u800cXGBoost\u5219\u53ef\u4ee5\u4eba\u4e3a\u5b9a\u4e49\u635f\u5931\u51fd\u6570\uff08\u53ef\u4ee5\u662f\u6700\u5c0f\u5e73\u65b9\u5dee\u3001logistic loss function\u3001hinge loss function\u6216\u8005\u4eba\u4e3a\u5b9a\u4e49\u7684loss function\uff09\uff0c\u53ea\u9700\u8981\u77e5\u9053\u8be5loss function\u5bf9\u53c2\u6570\u7684\u4e00\u9636\u3001\u4e8c\u9636\u5bfc\u6570\u4fbf\u53ef\u4ee5\u8fdb\u884cboosting\uff0c\u5176\u8fdb\u4e00\u6b65<span style=\"color: #ff0000;\"><strong>\u589e\u5927\u4e86\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b<\/strong><\/span>\uff0c\u5176\u8d2a\u5a6a\u6cd5\u5bfb\u627e\u6dfb\u52a0\u6811\u7684\u7ed3\u6784\u4ee5\u53caloss function\u4e2d\u7684\u635f\u5931\u51fd\u6570\u4e0e\u6b63\u5219\u9879\u7b49\u4e00\u7cfb\u5217\u7b56\u7565\u4e5f\u4f7f\u5f97XGBoost<span style=\"color: #ff0000;\"><strong>\u9884\u6d4b\u66f4\u51c6\u786e<\/strong><\/span>\u3002<\/p>\n<hr \/>\n<h6>## \u968f\u673a\u68ee\u6797<\/h6>\n<p><span style=\"color: #ff0000;\">\u968f\u673a\u68ee\u6797\u987e\u540d\u601d\u4e49\uff0c\u662f\u7528\u968f\u673a\u7684\u65b9\u5f0f\u5efa\u7acb\u4e00\u4e2a\u68ee\u6797\uff0c\u68ee\u6797\u91cc\u9762\u6709\u5f88\u591a\u7684\u51b3\u7b56\u6811\u7ec4\u6210\uff0c\u968f\u673a\u68ee\u6797\u7684\u6bcf\u4e00\u68f5\u51b3\u7b56\u6811\u4e4b\u95f4\u662f\u6ca1\u6709\u5173\u8054\u7684\u3002<\/span>\u5728\u5f97\u5230\u68ee\u6797\u4e4b\u540e\uff0c\u5f53\u6709\u4e00\u4e2a\u65b0\u7684\u8f93\u5165\u6837\u672c\u8fdb\u5165\u7684\u65f6\u5019\uff0c\u5c31\u8ba9\u68ee\u6797\u4e2d\u7684\u6bcf\u4e00\u68f5\u51b3\u7b56\u6811\u5206\u522b\u8fdb\u884c\u4e00\u4e0b\u5224\u65ad\uff0c\u770b\u770b\u8fd9\u4e2a\u6837\u672c\u5e94\u8be5\u5c5e\u4e8e\u54ea\u4e00\u7c7b\uff08\u5bf9\u4e8e\u5206\u7c7b\u7b97\u6cd5\uff09\uff0c\u7136\u540e\u770b\u770b\u54ea\u4e00\u7c7b\u88ab\u9009\u62e9\u6700\u591a\uff0c\u5c31\u9884\u6d4b\u8fd9\u4e2a\u6837\u672c\u4e3a\u90a3\u4e00\u7c7b\u3002<\/p>\n<p><strong>\u5728\u5efa\u7acb\u6bcf\u4e00\u68f5\u51b3\u7b56\u6811\u7684\u8fc7\u7a0b\u4e2d\uff0c\u6709\u4e24\u70b9\u9700\u8981\u6ce8\u610f &#8211; <span style=\"color: #ff0000;\">\u91c7\u6837<\/span>\u4e0e<span style=\"color: #ff0000;\">\u5b8c\u5168\u5206\u88c2<\/span>\u3002<\/strong>\u9996\u5148\u662f\u4e24\u4e2a\u968f\u673a\u91c7\u6837\u7684\u8fc7\u7a0b\uff0crandom forest\u5bf9\u8f93\u5165\u7684\u6570\u636e\u8981\u8fdb\u884c\u884c\u3001\u5217\u7684\u91c7\u6837\u3002\u5bf9\u4e8e\u884c\u91c7\u6837\uff0c\u91c7\u7528\u6709\u653e\u56de\u7684\u65b9\u5f0f\uff0c\u4e5f\u5c31\u662f\u5728\u91c7\u6837\u5f97\u5230\u7684\u6837\u672c\u96c6\u5408\u4e2d\uff0c\u53ef\u80fd\u6709\u91cd\u590d\u7684\u6837\u672c\u3002\u5047\u8bbe\u8f93\u5165\u6837\u672c\u4e3aN\u4e2a\uff0c\u90a3\u4e48\u91c7\u6837\u7684\u6837\u672c\u4e5f\u4e3aN\u4e2a\u3002\u8fd9\u6837\u4f7f\u5f97\u5728\u8bad\u7ec3\u7684\u65f6\u5019\uff0c\u6bcf\u4e00\u68f5\u6811\u7684\u8f93\u5165\u6837\u672c\u90fd\u4e0d\u662f\u5168\u90e8\u7684\u6837\u672c\uff0c\u4f7f\u5f97\u76f8\u5bf9\u4e0d\u5bb9\u6613\u51fa\u73b0over-fitting\u3002\u7136\u540e\u8fdb\u884c\u5217\u91c7\u6837\uff0c\u4eceM\u4e2afeature\u4e2d\uff0c\u9009\u62e9m\u4e2a(m &lt;&lt; M)\u3002\u4e4b\u540e\u5c31\u662f\u5bf9\u91c7\u6837\u4e4b\u540e\u7684\u6570\u636e\u4f7f\u7528\u5b8c\u5168\u5206\u88c2\u7684\u65b9\u5f0f\u5efa\u7acb\u51fa\u51b3\u7b56\u6811\uff0c\u8fd9\u6837\u51b3\u7b56\u6811\u7684\u67d0\u4e00\u4e2a\u53f6\u5b50\u8282\u70b9\u8981\u4e48\u662f\u65e0\u6cd5\u7ee7\u7eed\u5206\u88c2\u7684\uff0c\u8981\u4e48\u91cc\u9762\u7684\u6240\u6709\u6837\u672c\u7684\u90fd\u662f\u6307\u5411\u7684\u540c\u4e00\u4e2a\u5206\u7c7b\u3002\u4e00\u822c\u5f88\u591a\u7684\u51b3\u7b56\u6811\u7b97\u6cd5\u90fd\u4e00\u4e2a\u91cd\u8981\u7684\u6b65\u9aa4 &#8211; \u526a\u679d\uff0c\u4f46\u662f\u8fd9\u91cc\u4e0d\u8fd9\u6837\u5e72\uff0c\u7531\u4e8e\u4e4b\u524d\u7684\u4e24\u4e2a\u968f\u673a\u91c7\u6837\u7684\u8fc7\u7a0b\u4fdd\u8bc1\u4e86\u968f\u673a\u6027\uff0c\u6240\u4ee5\u5c31\u7b97\u4e0d\u526a\u679d\uff0c\u4e5f\u4e0d\u4f1a\u51fa\u73b0over-fitting\u3002<\/p>\n<p>\u6309\u8fd9\u79cd\u7b97\u6cd5\u5f97\u5230\u7684\u968f\u673a\u68ee\u6797\u4e2d\u7684\u6bcf\u4e00\u68f5\u90fd\u662f\u5f88\u5f31\u7684\uff0c\u4f46\u662f\u5927\u5bb6\u7ec4\u5408\u8d77\u6765\u5c31\u5f88\u5389\u5bb3\u4e86\u3002\u6211\u89c9\u5f97\u53ef\u4ee5\u8fd9\u6837\u6bd4\u55bb\u968f\u673a\u68ee\u6797\u7b97\u6cd5\uff1a\u6bcf\u4e00\u68f5\u51b3\u7b56\u6811\u5c31\u662f\u4e00\u4e2a\u7cbe\u901a\u4e8e\u67d0\u4e00\u4e2a\u7a84\u9886\u57df\u7684\u4e13\u5bb6\uff08\u56e0\u4e3a\u6211\u4eec\u4eceM\u4e2afeature\u4e2d\u9009\u62e9m\u8ba9\u6bcf\u4e00\u68f5\u51b3\u7b56\u6811\u8fdb\u884c\u5b66\u4e60\uff09\uff0c\u8fd9\u6837\u5728\u968f\u673a\u68ee\u6797\u4e2d\u5c31\u6709\u4e86\u5f88\u591a\u4e2a\u7cbe\u901a\u4e0d\u540c\u9886\u57df\u7684\u4e13\u5bb6\uff0c\u5bf9\u4e00\u4e2a\u65b0\u7684\u95ee\u9898\uff08\u65b0\u7684\u8f93\u5165\u6570\u636e\uff09\uff0c\u53ef\u4ee5\u7528\u4e0d\u540c\u7684\u89d2\u5ea6\u53bb\u770b\u5f85\u5b83\uff0c\u6700\u7ec8\u7531\u5404\u4e2a\u4e13\u5bb6\uff0c\u6295\u7968\u5f97\u5230\u7ed3\u679c\u3002<\/p>\n<h6>## GBDT(Gradient Boosting Decision Tree)<\/h6>\n<p>GBDT\u662f\u4e00\u4e2a\u5e94\u7528\u5f88\u5e7f\u6cdb\u7684\u7b97\u6cd5\uff0c\u53ef\u4ee5\u7528\u6765\u505a\u5206\u7c7b\u3001\u56de\u5f52\u3002\u5728\u5f88\u591a\u7684\u6570\u636e\u4e0a\u90fd\u6709\u4e0d\u9519\u7684\u6548\u679c\u3002GBDT\u8fd9\u4e2a\u7b97\u6cd5\u8fd8\u6709\u4e00\u4e9b\u5176\u4ed6\u7684\u540d\u5b57\uff0c\u6bd4\u5982\u8bf4MART(Multiple Additive Regression Tree)\uff0cGBRT(Gradient Boost Regression Tree)\uff0cTree Net\u7b49\uff0c\u5176\u5b9e\u5b83\u4eec\u90fd\u662f\u4e00\u4e2a\u4e1c\u897f\uff08\u53c2\u8003\u81eawikipedia \u2013 Gradient Boosting)\uff0c\u53d1\u660e\u8005\u662fFriedman\u3002<\/p>\n<p>Gradient Boost\u5176\u5b9e\u662f\u4e00\u4e2a\u6846\u67b6\uff0c\u91cc\u9762\u53ef\u4ee5\u5957\u5165\u5f88\u591a\u4e0d\u540c\u7684\u7b97\u6cd5\uff0c\u53ef\u4ee5\u53c2\u8003\u4e00\u4e0b\u673a\u5668\u5b66\u4e60\u4e0e\u6570\u5b66(3)\u4e2d\u7684\u8bb2\u89e3\u3002Boost\u662f&#8221;\u63d0\u5347&#8221;\u7684\u610f\u601d\uff0c\u4e00\u822cBoosting\u7b97\u6cd5\u90fd\u662f\u4e00\u4e2a\u8fed\u4ee3\u7684\u8fc7\u7a0b\uff0c\u6bcf\u4e00\u6b21\u65b0\u7684\u8bad\u7ec3\u90fd\u662f\u4e3a\u4e86\u6539\u8fdb\u4e0a\u4e00\u6b21\u7684\u7ed3\u679c\u3002<\/p>\n<p>\u539f\u59cb\u7684Boost\u7b97\u6cd5\u662f\u5728\u7b97\u6cd5\u5f00\u59cb\u7684\u65f6\u5019\uff0c\u4e3a\u6bcf\u4e00\u4e2a\u6837\u672c\u8d4b\u4e0a\u4e00\u4e2a\u6743\u91cd\u503c\uff0c\u521d\u59cb\u7684\u65f6\u5019\uff0c\u5927\u5bb6\u90fd\u662f\u4e00\u6837\u91cd\u8981\u7684\u3002\u5728\u6bcf\u4e00\u6b65\u8bad\u7ec3\u4e2d\u5f97\u5230\u7684\u6a21\u578b\uff0c\u4f1a\u4f7f\u5f97\u6570\u636e\u70b9\u7684\u4f30\u8ba1\u6709\u5bf9\u6709\u9519\uff0c\u6211\u4eec\u5c31\u5728\u6bcf\u4e00\u6b65\u7ed3\u675f\u540e\uff0c\u589e\u52a0\u5206\u9519\u7684\u70b9\u7684\u6743\u91cd\uff0c\u51cf\u5c11\u5206\u5bf9\u7684\u70b9\u7684\u6743\u91cd\uff0c\u8fd9\u6837\u4f7f\u5f97\u67d0\u4e9b\u70b9\u5982\u679c\u8001\u662f\u88ab\u5206\u9519\uff0c\u90a3\u4e48\u5c31\u4f1a\u88ab\u201c\u4e25\u91cd\u5173\u6ce8\u201d\uff0c\u4e5f\u5c31\u88ab\u8d4b\u4e0a\u4e00\u4e2a\u5f88\u9ad8\u7684\u6743\u91cd\u3002\u7136\u540e\u7b49\u8fdb\u884c\u4e86N\u6b21\u8fed\u4ee3\uff08\u7531\u7528\u6237\u6307\u5b9a\uff09\uff0c\u5c06\u4f1a\u5f97\u5230N\u4e2a\u7b80\u5355\u7684\u5206\u7c7b\u5668\uff08basic learner\uff09\uff0c\u7136\u540e\u6211\u4eec\u5c06\u5b83\u4eec\u7ec4\u5408\u8d77\u6765\uff08\u6bd4\u5982\u8bf4\u53ef\u4ee5\u5bf9\u5b83\u4eec\u8fdb\u884c\u52a0\u6743\u3001\u6216\u8005\u8ba9\u5b83\u4eec\u8fdb\u884c\u6295\u7968\u7b49\uff09\uff0c\u5f97\u5230\u4e00\u4e2a\u6700\u7ec8\u7684\u6a21\u578b\u3002<\/p>\n<p>\u800cGradient Boost\u4e0e\u4f20\u7edf\u7684Boost\u7684\u533a\u522b\u662f\uff0c\u6bcf\u4e00\u6b21\u7684\u8ba1\u7b97\u662f\u4e3a\u4e86\u51cf\u5c11\u4e0a\u4e00\u6b21\u7684\u6b8b\u5dee(residual)\uff0c\u800c\u4e3a\u4e86\u6d88\u9664\u6b8b\u5dee\uff0c\u6211\u4eec\u53ef\u4ee5\u5728\u6b8b\u5dee\u51cf\u5c11\u7684\u68af\u5ea6(Gradient)\u65b9\u5411\u4e0a\u5efa\u7acb\u4e00\u4e2a\u65b0\u7684\u6a21\u578b\u3002\u6240\u4ee5\u8bf4\uff0c\u5728Gradient Boost\u4e2d\uff0c\u6bcf\u4e2a\u65b0\u7684\u6a21\u578b\u7684\u5efa\u7acb\u662f\u4e3a\u4e86\u4f7f\u5f97\u4e4b\u524d\u6a21\u578b\u7684\u6b8b\u5dee\u5f80\u68af\u5ea6\u65b9\u5411\u51cf\u5c11\uff0c\u4e0e\u4f20\u7edfBoost\u5bf9\u6b63\u786e\u3001\u9519\u8bef\u7684\u6837\u672c\u8fdb\u884c\u52a0\u6743\u6709\u7740\u5f88\u5927\u7684\u533a\u522b\u3002<\/p>\n<p>&nbsp;<\/p>\n<h5 id=\"id-\u6a21\u677f-\u53c2\u8003\u94fe\u63a5\uff1a\">\u53c2\u8003\u94fe\u63a5\uff1a<\/h5>\n<ul>\n<li><a href=\"https:\/\/www.cnblogs.com\/LeftNotEasy\/archive\/2011\/01\/02\/machine-learning-boosting-and-gradient-boosting.html\">\u673a\u5668\u5b66\u4e60\u4e2d\u7684\u6570\u5b66(3)-\u6a21\u578b\u7ec4\u5408(Model Combining)\u4e4bBoosting\u4e0eGradient Boosting<\/a><\/li>\n<li><a href=\"https:\/\/www.cnblogs.com\/LeftNotEasy\/archive\/2011\/03\/07\/random-forest-and-gbdt.html\">\u673a\u5668\u5b66\u4e60\u4e2d\u7684\u7b97\u6cd5(1)-\u51b3\u7b56\u6811\u6a21\u578b\u7ec4\u5408\u4e4b\u968f\u673a\u68ee\u6797\u4e0eGBDT<\/a><\/li>\n<li><a href=\"https:\/\/blog.csdn.net\/v_JULY_v\/article\/details\/40718799\">Adaboost \u7b97\u6cd5\u7684\u539f\u7406\u4e0e\u63a8\u5bfc<\/a><\/li>\n<li><a href=\"https:\/\/zhuanlan.zhihu.com\/p\/42740654\">Adaboost\u3001GBDT\u4e0eXGBoost\u7684\u533a\u522b<\/a><\/li>\n<li><a href=\"https:\/\/wizardforcel.gitbooks.io\/dm-algo-top10\/content\/adaboost.html\">\u6570\u636e\u6316\u6398\u7b97\u6cd5\u5b66\u4e60\uff08\u516b\uff09Adaboost\u7b97\u6cd5<\/a><\/li>\n<li><a href=\"https:\/\/www.cnblogs.com\/pinard\/p\/6133937.html\">\u96c6\u6210\u5b66\u4e60\u4e4bAdaboost\u7b97\u6cd5\u539f\u7406\u5c0f\u7ed3<\/a><\/li>\n<li><a href=\"https:\/\/zhuanlan.zhihu.com\/p\/25096501\">\u5f53\u6211\u4eec\u5728\u8c08\u8bbaGBDT\uff1a\u4ece AdaBoost \u5230 Gradient Boosting<\/a><\/li>\n<li><a href=\"https:\/\/www.zhihu.com\/question\/54626685\">\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u4e2dGBDT\u4e0eAdaboost\u7684\u533a\u522b\u4e0e\u8054\u7cfb\u662f\u4ec0\u4e48\uff1f<\/a><\/li>\n<li><a href=\"https:\/\/blog.csdn.net\/HHTNAN\/article\/details\/80894247\">\u5173\u4e8eadaboost\u3001GBDT\u3001xgboost\u4e4b\u95f4\u7684\u533a\u522b\u4e0e\u8054\u7cfb<\/a><\/li>\n<li><a href=\"https:\/\/www.cnblogs.com\/willnote\/p\/6801496.html\">Boosting\u5b66\u4e60\u7b14\u8bb0\uff08Adboost\u3001GBDT\u3001Xgboost\uff09<\/a><\/li>\n<\/ul>\n<p>=END=<\/p>\n","protected":false},"excerpt":{"rendered":"<p>=Start= \u7f18\u7531\uff1a \u6700\u8fd1\u5728\u770b\u300a\u673a\u5668\u5b66\u4e60\u300b\u8fd9\u672c\u4e66\uff0c\u60f3\u7b80\u5355\u4e86\u89e3\u4e00\u4e0b\u673a\u5668\u5b66\u4e60\u539f\u7406\u3001\u7b97\u6cd5\u76f8\u5173\u7684\u5185\u5bb9\uff0c\u907f\u514d\u88ab\u65f6\u4ee3\u629b\u4e0b [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,1275],"tags":[1279,1278,1276,1280,9,1277],"class_list":["post-4242","post","type-post","status-publish","format-standard","hentry","category-knowledgebase-2","category-1275","tag-adaboost","tag-gbdt","tag-ml","tag-xgboost","tag-9","tag-1277"],"views":8424,"_links":{"self":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts\/4242","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/comments?post=4242"}],"version-history":[{"count":1,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts\/4242\/revisions"}],"predecessor-version":[{"id":4244,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts\/4242\/revisions\/4244"}],"wp:attachment":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/media?parent=4242"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/categories?post=4242"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/tags?post=4242"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}