{"id":305,"date":"2023-08-25T16:36:23","date_gmt":"2023-08-25T08:36:23","guid":{"rendered":"http:\/\/lingbo.online\/?p=305"},"modified":"2023-08-25T16:36:23","modified_gmt":"2023-08-25T08:36:23","slug":"catboost%e7%ae%97%e6%b3%95%e7%90%86%e8%ae%ba","status":"publish","type":"post","link":"https:\/\/lingbo.online\/index.php\/algorithm_learning\/catboost%e7%ae%97%e6%b3%95%e7%90%86%e8%ae%ba\/","title":{"rendered":"CatBoost\u7b97\u6cd5\u7406\u8bba"},"content":{"rendered":"<h3>\u6982\u8ff0<\/h3>\n<p>CatBoost\u662f\u4fc4\u7f57\u65af\u7684\u641c\u7d22\u5de8\u5934Yandex\u57282017\u5e74\u5f00\u6e90\u7684\u673a\u5668\u5b66\u4e60\u5e93\uff0c\u662fBoosting\u65cf\u7b97\u6cd5\u7684\u4e00\u79cd\u3002CatBoost\u548cXGBoost\u3001LightGBM\u5e76\u79f0\u4e3aGBDT\u7684\u4e09\u5927\u4e3b\u6d41\u795e\u5668\uff0c\u90fd\u662f\u5728GBDT\u7b97\u6cd5\u6846\u67b6\u4e0b\u7684\u4e00\u79cd\u6539\u8fdb\u5b9e\u73b0\u3002XGBoost\u88ab\u5e7f\u6cdb\u7684\u5e94\u7528\u4e8e\u5de5\u4e1a\u754c\uff0cLightGBM\u6709\u6548\u7684\u63d0\u5347\u4e86GBDT\u7684\u8ba1\u7b97\u6548\u7387\uff0c\u800cYandex\u7684CatBoost\u53f7\u79f0\u662f\u6bd4XGBoost\u548cLightGBM\u5728\u7b97\u6cd5\u51c6\u786e\u7387\u7b49\u65b9\u9762\u8868\u73b0\u66f4\u4e3a\u4f18\u79c0\u7684\u7b97\u6cd5\u3002<\/p>\n<p>CatBoost\u662f\u4e00\u79cd\u57fa\u4e8e\u5bf9\u79f0\u51b3\u7b56\u6811\uff08oblivious trees\uff09\u4e3a\u57fa\u5b66\u4e60\u5668\u5b9e\u73b0\u7684\u53c2\u6570\u8f83\u5c11\u3001\u652f\u6301\u7c7b\u522b\u578b\u53d8\u91cf\u548c\u9ad8\u51c6\u786e\u6027\u7684GBDT\u6846\u67b6\uff0c\u4e3b\u8981\u89e3\u51b3\u7684\u75db\u70b9\u662f\u9ad8\u6548\u5408\u7406\u5730\u5904\u7406\u7c7b\u522b\u578b\u7279\u5f81\uff0cCatBoost\u662f\u7531Categorical\u548cBoosting\u7ec4\u6210\u3002\u6b64\u5916\uff0cCatBoost\u8fd8\u89e3\u51b3\u4e86\u68af\u5ea6\u504f\u5dee\uff08Gradient Bias\uff09\u53ca\u9884\u6d4b\u504f\u79fb\uff08Prediction shift\uff09\u7684\u95ee\u9898\uff0c\u4ece\u800c\u51cf\u5c11\u8fc7\u62df\u5408\u7684\u53d1\u751f\uff0c\u8fdb\u800c\u63d0\u9ad8\u7b97\u6cd5\u7684\u51c6\u786e\u6027\u548c\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n<h3>\u7b97\u6cd5\u7279\u70b9<\/h3>\n<ul>\n<li>Catboost\u81ea\u52a8\u91c7\u7528\u7279\u6b8a\u7684\u65b9\u5f0f\u5904\u7406\u7c7b\u522b\u578b\u7279\u5f81\uff08categorical features\uff09\u3002\u9996\u5148\u5bf9categorical features\u505a\u4e00\u4e9b\u7edf\u8ba1\uff0c\u8ba1\u7b97\u67d0\u4e2a\u7c7b\u522b\u7279\u5f81\uff08category\uff09\u51fa\u73b0\u7684\u9891\u7387\uff0c\u4e4b\u540e\u52a0\u4e0a\u8d85\u53c2\u6570\uff0c\u751f\u6210\u65b0\u7684\u6570\u503c\u578b\u7279\u5f81\uff08numerical features\uff09\u3002Catboost\u901a\u8fc7\u8be5\u65b9\u6cd5\u63d0\u4f9b\u4e86\u81ea\u52a8\u5904\u7406\u7c7b\u522b\u7279\u5f81\u7684\u80fd\u529b\uff0c\u89e3\u51b3\u4e86\u7b97\u6cd5\u5de5\u7a0b\u9700\u8981\u5bf9\u7c7b\u522b\u7279\u5f81\u8fdb\u884c\u7279\u6b8a\u5904\u7406\u7684\u75db\u70b9\u95ee\u9898\u3002<\/li>\n<li>Catboost\u4f7f\u7528\u7ec4\u5408\u7c7b\u522b\u7279\u5f81\uff0c\u53ef\u4ee5\u5229\u7528\u5230\u7279\u5f81\u4e4b\u95f4\u7684\u8054\u7cfb\uff0c\u4ece\u800c\u6781\u5927\u7684\u4e30\u5bcc\u4e86\u7279\u5f81\u7ef4\u5ea6\u3002<\/li>\n<li>Catboost\u91c7\u7528ordered boost\u7684\u65b9\u6cd5\u907f\u514d\u68af\u5ea6\u4f30\u8ba1\u7684\u504f\u5dee\uff0c\u8fdb\u800c\u89e3\u51b3\u9884\u6d4b\u504f\u79fb\u7684\u95ee\u9898\u3002<\/li>\n<li>Catboost\u7684\u57fa\u6a21\u578b\u91c7\u7528\u7684\u662f\u5bf9\u79f0\u6811\uff0c\u540c\u65f6\u8ba1\u7b97leaf-value\u65b9\u5f0f\u548c\u4f20\u7edf\u7684boosting\u7b97\u6cd5\u4e5f\u4e0d\u4e00\u6837\uff0c\u4f20\u7edf\u7684boosting\u7b97\u6cd5\u8ba1\u7b97\u7684\u662f\u5e73\u5747\u6570\uff0c\u800ccatboost\u5728\u8fd9\u65b9\u9762\u505a\u4e86\u4f18\u5316\u91c7\u7528\u4e86\u5176\u4ed6\u7684\u7b97\u6cd5\uff0c\u8fd9\u4e9b\u6539\u8fdb\u90fd\u80fd\u9632\u6b62\u6a21\u578b\u8fc7\u62df\u5408\u3002<\/li>\n<\/ul>\n<h4>Catboost\u4f18\u7f3a\u70b9<\/h4>\n<h5>Catboost\u4f18\u70b9<\/h5>\n<ul>\n<li>\u6027\u80fd\u5353\u8d8a\uff1a\u5728\u6027\u80fd\u65b9\u9762\u53ef\u4ee5\u5339\u654c\u4efb\u4f55\u5148\u8fdb\u7684\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\uff1b<\/li>\n<li>\u9c81\u68d2\u6027\/\u5065\u58ee\u6027\uff1a\u5b83\u51cf\u5c11\u4e86\u5bf9\u5f88\u591a\u8d85\u53c2\u8c03\u4f18\u7684\u9700\u6c42\uff0c\u5e76\u4e14\u964d\u4f4e\u4e86\u8fc7\u62df\u5408\u7684\u6982\u7387\uff0c\u589e\u5f3a\u4e86\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\uff1b<\/li>\n<li>\u6613\u4e8e\u4f7f\u7528\uff1a\u63d0\u4f9b\u4e0escikit\u96c6\u6210\u7684Python\u63a5\u53e3\uff0c\u4ee5\u53caR\u548c\u547d\u4ee4\u884c\u754c\u9762\uff1b<\/li>\n<li>\u5b9e\u7528\uff1a\u53ef\u4ee5\u5904\u7406\u7c7b\u522b\u578b\u3001\u6570\u503c\u578b\u7279\u5f81\uff1b<\/li>\n<li>\u53ef\u62d3\u5c55\uff1a\u652f\u6301\u81ea\u5b9a\u4e49\u635f\u5931\u51fd\u6570\uff1b<\/li>\n<li>\u652f\u6301\u7c7b\u522b\u578b\u53d8\u91cf\uff0c\u65e0\u9700\u5bf9\u975e\u6570\u503c\u578b\u7279\u5f81\u8fdb\u884c\u9884\u5904\u7406\uff1b<\/li>\n<li>\u5feb\u901f\u3001\u53ef\u62d3\u5c55\u7684GPU\u7248\u672c\uff0c\u53ef\u4ee5\u4f7f\u7528\u57fa\u4e8eGPU\u7684\u68af\u5ea6\u63d0\u5347\u7b97\u6cd5\u8bad\u7ec3\u6a21\u578b\uff0c\u652f\u6301\u591a\u5361\u5e76\u884c\uff1b<\/li>\n<li>\u5feb\u901f\u9884\u6d4b\uff0c\u5bf9\u5ef6\u65f6\u975e\u5e38\u82db\u523b\u7684\u4efb\u52a1\u4e5f\u80fd\u5feb\u901f\u9ad8\u6548\u7684\u90e8\u7f72\u6a21\u578b\u3002<\/li>\n<\/ul>\n<h5>Catboost\u7f3a\u70b9<\/h5>\n<ul>\n<li>\u5bf9\u4e8e\u7c7b\u522b\u578b\u7279\u5f81\u7684\u5904\u7406\u9700\u8981\u5927\u91cf\u7684\u5185\u5b58\u548c\u65f6\u95f4\uff1b<\/li>\n<li>\u4e0d\u540c\u968f\u673a\u6570\u7684\u8bbe\u5b9a\u5bf9\u4e8e\u6a21\u578b\u9884\u6d4b\u7ed3\u679c\u6709\u4e00\u5b9a\u7684\u5f71\u54cd\uff1b<\/li>\n<\/ul>\n<p>\u4e0eXGBoost\u3001LightGBM\u76f8\u6bd4\uff0cCatBoost\u7684\u521b\u65b0\u70b9\u6709\uff1a<\/p>\n<h5>CatBoost\u521b\u65b0\u70b9<\/h5>\n<ul>\n<li>\u5d4c\u5165\u4e86\u81ea\u52a8\u5c06\u7c7b\u522b\u7279\u5f81\u5904\u7406\u4e3a\u6570\u503c\u578b\u7279\u5f81\u7684\u521b\u65b0\u7b97\u6cd5\u3002\u9996\u5148\u5bf9categorical features\u8fdb\u884c\u7edf\u8ba1\uff0c\u8ba1\u7b97\u67d0\u4e2a\u7c7b\u522b\u7279\u5f81\uff08category\uff09\u51fa\u73b0\u7684\u9891\u7387\uff0c\u4e4b\u540e\u52a0\u4e0a\u8d85\u53c2\u6570\uff0c\u751f\u6210\u65b0\u7684\u6570\u503c\u578b\u7279\u5f81\uff08numerical features\uff09\uff1b<\/li>\n<li>Catboost\u4f7f\u7528\u7ec4\u5408\u7c7b\u522b\u7279\u5f81\uff0c\u53ef\u4ee5\u5229\u7528\u5230\u7279\u5f81\u4e4b\u95f4\u7684\u8054\u7cfb\uff0c\u6781\u5927\u4e30\u5bcc\u4e86\u7279\u5f81\u7ef4\u5ea6\uff1b<\/li>\n<li>\u91c7\u7528\u6392\u5e8f\u63d0\u5347\u7684\u65b9\u6cd5\u5bf9\u6297\u8bad\u7ec3\u96c6\u4e2d\u7684\u566a\u58f0\u70b9\uff0c\u4ece\u800c\u907f\u514d\u68af\u5ea6\u4f30\u8ba1\u7684\u504f\u5dee\uff0c\u8fdb\u800c\u89e3\u51b3\u9884\u6d4b\u504f\u79fb\u7684\u95ee\u9898\uff1b<\/li>\n<li>\u91c7\u7528\u5b8c\u5168\u5bf9\u79f0\u6811\u4f5c\u4e3a\u57fa\u6a21\u578b\uff1b<\/li>\n<li>\u91c7\u7528\u4e00\u79cd\u65b0\u7684\u7b97\u6cd5\u8ba1\u7b97leaf-values\u3002<\/li>\n<\/ul>\n<h3>\u7b97\u6cd5\u539f\u7406<\/h3>\n<h4>Ordered Target Statistics<\/h4>\n<p>\u5bf9\u4e8e\u7c7b\u522b\u7279\u5f81\uff0c\u5982\u679c\u7c7b\u522b\u6570\u76ee\u4e0d\u591a\uff0c\u53ef\u4ee5\u4f7f\u7528onehot\u7f16\u7801\u3002\u5426\u5219\uff0c\u5f88\u5bb9\u6613\u9020\u6210\u7ef4\u5ea6\u7206\u70b8\u3002CatBoost\u4e0d\u63d0\u5021\u4f7f\u7528one-hot\u7f16\u7801\uff0c\u5b83\u8bbe\u8ba1\u4e86\u4e00\u79cd\u57fa\u4e8e\u9884\u6d4b\u76ee\u6807\u7edf\u8ba1\u503c\u7684\u65b9\u6cd5\u53ef\u4ee5\u5c06\u7c7b\u522b\u7279\u5f81\u8f6c\u5316\u4e3a\u6570\u503c\u7279\u5f81\u3002\u9887\u6709\u5747\u503c\u7f16\u7801\u7684\u601d\u60f3\u3002<\/p>\n<p>CatBoost\u7b97\u6cd5\u7684\u8bbe\u8ba1\u521d\u8877\u662f\u4e3a\u4e86\u66f4\u597d\u7684\u5904\u7406GBDT\u7279\u5f81\u4e2d\u7684categorical features\u3002\u5728\u5904\u7406GBDT\u7279\u5f81\u4e2d\u7684categorical features\u7684\u65f6\u5019\uff0c\u6700\u7b80\u5355\u7684\u65b9\u6cd5\u662f\u7528 categorical feature\u5bf9\u5e94\u7684\u6807\u7b7e\u7684\u5e73\u5747\u503c\u6765\u66ff\u6362\u3002\u5728\u51b3\u7b56\u6811\u4e2d\uff0c\u6807\u7b7e\u5e73\u5747\u503c\u5c06\u4f5c\u4e3a\u8282\u70b9\u5206\u88c2\u7684\u6807\u51c6\u3002\u8fd9\u79cd\u65b9\u6cd5\u88ab\u79f0\u4e3a Greedy Target-based Statistics , \u7b80\u79f0 Greedy TS\uff0c\u7528\u516c\u5f0f\u6765\u8868\u8fbe\u5c31\u662f\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/lingbo.online\/wp-content\/uploads\/2023\/08\/greedy_ts-1.png\" alt=\"\" \/><\/p>\n<p>\u7136\u800c\uff0c\u8fd9\u79cd\u65b9\u6cd5\u6709\u4e00\u4e2a\u663e\u800c\u6613\u89c1\u7684\u7f3a\u9677\uff0c\u5373\u901a\u5e38\u7279\u5f81\u6bd4\u6807\u7b7e\u5305\u542b\u66f4\u591a\u7684\u4fe1\u606f\uff0c\u5982\u679c\u5f3a\u884c\u4f7f\u7528\u6807\u7b7e\u7684\u5e73\u5747\u503c\u6765\u8868\u793a\u7279\u5f81\u7684\u8bdd\uff0c\u5f53\u8bad\u7ec3\u6570\u636e\u96c6\u548c\u6d4b\u8bd5\u6570\u636e\u96c6\u6570\u636e\u7ed3\u6784\u548c\u5206\u5e03\u4e0d\u4e00\u6837\u7684\u65f6\u5019\u4f1a\u51fa\u6761\u4ef6\u504f\u79fb\u95ee\u9898\u3002<\/p>\n<p>CatBoost \u4f7f\u7528\u4e00\u4e2a\u66f4\u6709\u6548\u7684\u7b56\u7565\uff0c\u4e00\u65b9\u9762\u53ef\u4ee5\u51cf\u5c11\u8fc7\u62df\u5408\uff0c\u53e6\u4e00\u65b9\u9762\u4f7f\u7528\u5168\u90e8\u6570\u636e\u6765\u8bad\u7ec3\u3002\u5bf9\u6570\u636e\u96c6\u5148\u968f\u673a\u6392\u5e8f\uff0c\u5bf9\u4e8e\u6bcf\u4e2a\u6837\u672c\u7684\u8be5\u7c7b\u522b\u7279\u5f81\u4e2d\u7684\u67d0\u4e2a\u53d6\u503c\uff0c\u8f6c\u6362\u4e3a\u6570\u503c\u578b\u65f6\u90fd\u662f\u57fa\u4e8e\u8be5\u6837\u672c\u4e4b\u524d\u7684\u7c7b\u522blabel value\u53d6\u5747\u503c\uff0c\u540c\u65f6\u52a0\u5165\u4e86\u4f18\u5148\u7ea7\uff08\u5148\u9a8c\u503c\uff09\u7684\u6743\u91cd\u7cfb\u6570\u3002<\/p>\n<p>\u5047\u8bbe<span class=\"katex-eq\" data-katex-display=\"false\">\/sigma=(\/sigma_1,\/sigma_2,...,\/sigma_n)<\/span>\u662f\u968f\u673a\u6392\u5217\u5e8f\u5217\uff0c\u5219\u6709\uff1a<br \/>\n<img decoding=\"async\" src=\"https:\/\/lingbo.online\/wp-content\/uploads\/2023\/08\/ts_order.png\" alt=\"\" \/><br \/>\n[\u00b7]\u4ee3\u8868\u6307\u793a\u51fd\u6570\uff0cP\u5c31\u4ee3\u8868\u5148\u9a8c\uff0c\u5bf9\u5e94\u56de\u5f52\u4efb\u52a1\uff0c\u8ba1\u7b97\u6807\u7b7e\u7684\u5e73\u5747\u503c\u4e3a\u5148\u9a8c\u503c\uff1b\u5bf9\u4e8e\u4e8c\u5206\u4efb\u52a1\uff0c\u5c06\u6b63\u7c7b\u7684\u51fa\u73b0\u6982\u7387\u4f5c\u4e3a\u5148\u9a8c\u503c\u3002<span class=\"katex-eq\" data-katex-display=\"false\">\/alpha<\/span>\u5c31\u4ee3\u8868\u4f18\u5148\u7ea7\u7684\u6743\u91cd\u7cfb\u6570\uff0c\u7528\u6765\u9632\u6b62\u4f4e\u9891\u6b21\u7279\u5f81\u5e26\u6765\u7684\u5f71\u54cd\u6240\u6709\u7684\u5e73\u6ed1\u64cd\u4f5c\uff0c\u5982\u679c\u4e0d\u4f7f\u7528\u8fd9\u4e2a\u64cd\u4f5c\u7684\u8bdd\uff0c\u5f53\u5bf9\u4e8e\u67d0\u4e00\u4e2a\u7279\u5f81\u53ea\u6709\u4e00\u4e2a\u6837\u672c\u7684\u65f6\u5019\uff0c\u5176\u7279\u5f81\u7f16\u7801\u5c31\u662f1\uff0c\u4f1a\u6709\u8fc7\u62df\u5408\u7684\u98ce\u9669\u3002<br \/>\n\u8fd9\u79cd\u7c7b\u522b\u6807\u7b7e\u5904\u7406\u65b9\u6cd5\u5c31\u662f<strong>Ordered Target Statistics<\/strong>\u6570\u503c\u7f16\u7801\u65b9\u6cd5\uff0c\u53ef\u4ee5\u6709\u6548\u89e3\u51b3\u9884\u6d4b\u6f02\u79fb\u7684\u95ee\u9898\u3002<\/p>\n<h4>\u57fa\u4e8e\u8d2a\u5fc3\u7b56\u7565\u7684\u7279\u5f81\u4ea4\u53c9\u65b9\u6cd5<\/h4>\n<p>\u4f7f\u7528Ordered Target Statistics\u65b9\u6cd5\u5c06\u7c7b\u522b\u7279\u5f81\u8f6c\u5316\u6210\u4e3a\u6570\u503c\u7279\u5f81\u4ee5\u540e\uff0c\u4f1a\u5f71\u54cd\u5230\u7279\u5f81\u4ea4\u53c9\uff0c\u56e0\u4e3a\u6570\u503c\u7279\u5f81\u65e0\u6cd5\u6709\u6548\u5730\u8fdb\u884c\u4ea4\u53c9\u3002\u4e3a\u4e86\u6709\u6548\u5730\u5229\u7528\u7279\u5f81\u4ea4\u53c9\uff0cCatBoost\u5728\u5c06\u7c7b\u522b\u7279\u5f81\u8f6c\u6362\u4e3a\u6570\u503c\u7f16\u7801\u7684\u540c\u65f6\uff0c\u4f1a\u81ea\u52a8\u751f\u6210\u4ea4\u53c9\u7279\u5f81\u3002\u4f46\u5982\u679c\u8ba9\u5168\u90e8\u7684\u7c7b\u522b\u7279\u5f81\u4e4b\u95f4\u90fd\u8fdb\u884c\u4ea4\u53c9\uff0c\u4e24\u4e24\u4ea4\u53c9\uff0c\u4e09\u4e09\u4ea4\u53c9\uff0c\u56db\u56db\u4ea4\u53c9\uff0c\u8fd9\u4e2a\u590d\u6742\u5ea6\u662f\u6307\u6570\u7ea7\u7684\uff0c\u7279\u5f81\u7ef4\u5ea6\u4e00\u5b9a\u4f1a\u7206\u70b8\u3002<\/p>\n<p>CatBoost\u4f7f\u7528\u4e00\u79cd\u8d2a\u5fc3\u7684\u7b56\u7565\u6765\u8fdb\u884c\u7279\u5f81\u4ea4\u53c9\u3002\u751f\u6210tree\u7684\u7b2c\u4e00\u6b21\u5206\u88c2\uff0cCatBoost\u4e0d\u4f7f\u7528\u4efb\u4f55\u4ea4\u53c9\u7279\u5f81\u3002\u5728\u540e\u9762\u7684\u5206\u88c2\u4e2d\uff0cCatBoost\u4f1a\u4f7f\u7528\u751f\u6210tree\u6240\u7528\u5230\u7684\u5168\u90e8\u539f\u59cb\u7279\u5f81\u548c\u4ea4\u53c9\u7279\u5f81\u8ddf\u6570\u636e\u96c6\u4e2d\u7684\u5168\u90e8\u7c7b\u522b\u7279\u5f81\u8fdb\u884c\u4ea4\u53c9\u3002<\/p>\n<p>\u53c2\u6570<code>max_ctr_complexity<\/code>\u7528\u4e8e\u63a7\u5236\u7279\u5f81\u4ea4\u53c9\u7684\u6700\u5927\u4e2a\u6570\u3002<\/p>\n<h4>\u907f\u514d\u9884\u6d4b\u504f\u79fb\u7684Ordered Boosting\u65b9\u6cd5<\/h4>\n<p>\u4f7f\u7528XGBoost\u6216\u8005LightGBM\u505a\u6a21\u578b\u65f6\uff0c\u53ef\u80fd\u4f1a\u51fa\u73b0\u6a21\u578b\u5728\u8bad\u7ec3\u96c6\u4e0a\u62df\u5408\u7684\u5f88\u597d\uff0ctrain_auc\u751a\u81f3\u8fbe\u5230\u4e861.0\uff1b\u4f46\u662f\u5728\u9a8c\u8bc1\u96c6\u4e0a\u53d8\u5dee\u7684\u95ee\u9898, val_auc \u53ef\u80fd\u53ea\u67090.7\u3002\u8fd9\u5f53\u7136\u6709\u53ef\u80fd\u662f\u56e0\u4e3atree\u7684\u6570\u91cf\u592a\u591a\u4e86\uff0c\u6216\u8005\u662f\u6bcf\u68f5tree\u7684leaves\u592a\u591a\u4e86\uff0c\u5373\u6a21\u578b\u592a\u590d\u6742\u4e86\u9020\u6210\u4e86\u8fc7\u62df\u5408\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u8fd9\u4e5f\u53ef\u80fd\u662f\u7531\u4e8eXGBoost\u548cLightGBM\u81ea\u8eab\u7b97\u6cd5\u7684\u7f3a\u9677\u56e0\u7d20\u5bfc\u81f4\u3002LightGBM\u5728\u8bad\u7ec3\u4e0b\u4e00\u68f5tree\u7684\u65f6\u5019\uff0c\u9700\u8981\u8ba1\u7b97\u524d\u9762\u8fd9\u4e9btree\u6784\u6210\u7684\u52a0\u6cd5\u6a21\u578b\u5728\u6240\u6709\u6837\u672c\u4e0a\u7684\u4e00\u9636\u68af\u5ea6\u548c\u4e8c\u9636\u68af\u5ea6(Loss\u5bf9\u6a21\u578b\u9884\u6d4b\u7ed3\u679c\u7684\u5bfc\u6570)\uff0c\u7136\u540e\u7528\u8fd9\u4e9b\u68af\u5ea6\u6765\u51b3\u5b9a\u4e0b\u4e00\u68f5\u6811\u7684\u7ed3\u6784\u548c\u53f6\u5b50\u8282\u70b9\u53d6\u503c\u3002<\/p>\n<p>\u7136\u800c\uff0c\u8ba1\u7b97\u4e00\u9636\u68af\u5ea6\u548c\u4e8c\u9636\u68af\u5ea6\u503c\u662f\u6709\u95ee\u9898\u7684\u3002\u524d\u9762\u7684\u8fd9\u4e9btree\u90fd\u662f\u5728\u8fd9\u4e9b\u6837\u672c\u4e0a\u8bad\u7ec3\u7684\uff0c\u73b0\u5728\u53c8\u5728\u8fd9\u4e9b\u6837\u672c\u4e0a\u4f30\u8ba1\u6a21\u578b\u9884\u6d4b\u7ed3\u679c\u7684\u4e00\u9636\u548c\u4e8c\u9636\u68af\u5ea6\u3002\u663e\u7136\uff0c\u5e94\u8be5\u6362\u4e00\u4e9b\u65b0\u7684\u6837\u672c\u624d\u66f4\u5408\u7406\u3002<\/p>\n<p>\u5728CatBoost\u4e2d\uff0c\u5148\u5c06\u6837\u672c\u968f\u673a\u6253\u4e71\uff0c\u7136\u540e\u6bcf\u4e2a\u6837\u672c\u53ea\u4f7f\u7528\u6392\u5e8f\u5728\u5b83\u524d\u9762\u7684\u6837\u672c\u6765\u8bad\u7ec3\u6a21\u578b\u3002\u7528\u8fd9\u6837\u7684\u6a21\u578b\u6765\u4f30\u8ba1\u8fd9\u4e2a\u6837\u672c\u9884\u6d4b\u7ed3\u679c\u7684\u4e00\u9636\u548c\u4e8c\u9636\u68af\u5ea6\u3002\u7136\u540e\u7528\u8fd9\u4e9b\u68af\u5ea6\u6784\u5efa\u4e00\u68f5tree\u7684\u7ed3\u6784\uff0c\u6700\u7ec8tree\u7684\u6bcf\u4e2a\u53f6\u5b50\u8282\u70b9\u7684\u53d6\u503c\uff0c\u662f\u4f7f\u7528\u5168\u4f53\u6837\u672c\u8fdb\u884c\u8ba1\u7b97\u7684\u3002<\/p>\n<p>\u8fd9\u5c31\u662fOrdered Boosting\u7684\u4e3b\u8981\u601d\u60f3\u3002\u53ef\u4ee5\u6709\u6548\u5730\u51cf\u5c11\u68af\u5ea6\u4f30\u8ba1\u7684\u8bef\u5dee\uff0c\u7f13\u89e3\u9884\u6d4b\u504f\u79fb\u3002\u4f46\u662f\u4f1a\u589e\u52a0\u8f83\u591a\u7684\u8ba1\u7b97\u91cf\uff0c\u5f71\u54cd\u8bad\u7ec3\u901f\u5ea6\u3002<\/p>\n<p>\u5728\u5b9a\u4e49CatBoost\u6a21\u578b\u65f6\uff0c\u53ef\u4ee5\u7528<code>boosting_type<\/code>\u8fd9\u4e2a\u53c2\u6570\u6765\u8bbe\u7f6e\u662f\u4f7f\u7528Ordered Boosting\u8fd8\u662fLightGBM\u90a3\u6837\u7684Plain Boosting\u3002\u5982\u679c\u4e0d\u663e\u5f0f\u8bbe\u7f6e\uff0cCatBoost\u4f1a\u6839\u636e\u6837\u672c\u548c\u7279\u5f81\u6570\u91cf\u81ea\u884c\u51b3\u5b9a\u3002<\/p>\n<h5>\u8be6\u7ec6\u6b65\u9aa4<\/h5>\n<p>\u6784\u5efa\u4e00\u68f5\u6811\u6709\u4e24\u4e2a\u9636\u6bb5\uff1a\u7b2c\u4e00\uff0c\u9009\u62e9\u6811\u7ed3\u6784\uff1b\u7b2c\u4e8c\uff0c\u5728\u6811\u7ed3\u6784\u56fa\u5b9a\u540e\u8ba1\u7b97\u53f6\u8282\u70b9\u7684\u503c\u3002CatBoost\u5728\u7b2c\u4e8c\u9636\u6bb5\u91c7\u7528\u4f20\u7edfGBDT\u65b9\u6cd5\u6267\u884c\uff0c\u5728\u7b2c\u4e00\u9636\u6bb5\u5219\u91c7\u7528\u4fee\u6b63\u7684\u65b9\u6cd5\u2014\u2014\u68af\u5ea6\u6b65\u957f\u7684\u65e0\u504f\u4f30\u8ba1\u3002<\/p>\n<p>\u4ee4<span class=\"katex-eq\" data-katex-display=\"false\">F^i<\/span>\u4e3a\u524d<span class=\"katex-eq\" data-katex-display=\"false\">i<\/span>\u68f5\u6811\u7684\u6a21\u578b\u7ed3\u6784\uff0c\u4e3a\u4e86\u4f7f<span class=\"katex-eq\" data-katex-display=\"false\">g^i(X_k,Y_k)<\/span>\u662f\u5173\u4e8e\u6a21\u578b<span class=\"katex-eq\" data-katex-display=\"false\">F^i<\/span>\u7684\u65e0\u504f\u68af\u5ea6\uff0c\u9700\u8981\u5728\u8bad\u7ec3\u7684\u65f6\u5019\u4e0d\u4f7f\u7528\u6837\u672ck\u3002<br \/>\n\u6837\u672c\u96c6\u4e3a<span class=\"katex-eq\" data-katex-display=\"false\">\\lbrace(X_k, Y_k)\\rbrace^{n}_{k=1}<\/span>\u6309\u968f\u673a\u6392\u5e8f<span class=\"katex-eq\" data-katex-display=\"false\">\\sigma<\/span>\u6392\u5e8f\uff0c\u6811\u7684\u68f5\u6811\u4e3a<span class=\"katex-eq\" data-katex-display=\"false\">I<\/span>\u3002<\/p>\n<ul>\n<li>\u9996\u5148\uff0c\u5bf9\u4e8e\u6837\u672c<span class=\"katex-eq\" data-katex-display=\"false\">X_k<\/span>,\u521d\u59cb\u5316\u6a21\u578b<span class=\"katex-eq\" data-katex-display=\"false\">M_k<\/span>\uff1b<\/li>\n<li>\u5bf9\u4e8e\u6bcf\u4e00\u68f5\u6811\uff0c\u904d\u5386\u4e00\u6b21\u6837\u672c\uff0c\u5bf9\u4e8e\u524d<span class=\"katex-eq\" data-katex-display=\"false\">k-1<\/span>\u4e2a\u6837\u672c\uff0c\u4e00\u4f9d\u6b21\u8ba1\u7b97Loss\u7684\u68af\u5ea6<span class=\"katex-eq\" data-katex-display=\"false\">g^i<\/span>;<\/li>\n<li>\u518d\u6b21\u5c06\u524d<span class=\"katex-eq\" data-katex-display=\"false\">k-1<\/span>\u4e2a\u6837\u672c\u7684<span class=\"katex-eq\" data-katex-display=\"false\">g^i<\/span>\u548c<span class=\"katex-eq\" data-katex-display=\"false\">X^j(j=1,\u2026,k-1)<\/span>\u6765\u6784\u5efa\u6a21\u578b<span class=\"katex-eq\" data-katex-display=\"false\">M<\/span>\uff1b<\/li>\n<li>\u6700\u540e\uff0c\u5bf9\u4e8e\u6bcf\u4e00\u4e2a\u6837\u672c<span class=\"katex-eq\" data-katex-display=\"false\">X_k<\/span>\uff0c\u7528<span class=\"katex-eq\" data-katex-display=\"false\">M<\/span>\u6765\u4fee\u6b63\u521d\u59cb\u5316\u7684<span class=\"katex-eq\" data-katex-display=\"false\">M_k<\/span>\uff0c\u8fd9\u6837\u5c31\u53ef\u4ee5\u5f97\u5230\u4e00\u4e2a\u5206\u9694\u7684\u6a21\u578b<span class=\"katex-eq\" data-katex-display=\"false\">M_k<\/span>\uff08\u5e76\u4e14\u8fd9\u4e2a\u6a21\u578b\u4e0d\u9700\u8981\u8fd9\u4e2a\u6837\u672c\u7528\u68af\u5ea6\u4f30\u8ba1\u6765\u66f4\u65b0\uff09\u3002\u91cd\u590d\u4e0a\u8ff0\u64cd\u4f5c\uff0c\u5c31\u53ef\u4ee5\u5f97\u5230\u6bcf\u4e2a\u6837\u672c<span class=\"katex-eq\" data-katex-display=\"false\">X<\/span>\u7684\u5206\u9694\u6a21\u578b<span class=\"katex-eq\" data-katex-display=\"false\">M<\/span>\u3002\u7531\u6b64\u53ef\u89c1\uff0c\u6bcf\u4e00\u4e2a<span class=\"katex-eq\" data-katex-display=\"false\">M_k<\/span>\u90fd\u5171\u4eab\u76f8\u540c\u7684\u6811\u7ed3\u6784\u3002<\/li>\n<\/ul>\n<p><img decoding=\"async\" src=\"https:\/\/lingbo.online\/wp-content\/uploads\/2023\/08\/Inkedoder.jpg\" alt=\"\" \/><\/p>\n<h4>\u4f7f\u7528\u5bf9\u79f0\u4e8c\u53c9\u6811\u4f5c\u4e3a\u57fa\u6a21\u578b<\/h4>\n<p>XGBoost\u548cLightGBM\u91c7\u7528\u7684\u57fa\u6a21\u578b\u662f\u666e\u901a\u7684\u4e8c\u53c9\u6811\uff0c\u4f46\u662fCatBoost\u91c7\u7528\u7684\u662f\u5bf9\u79f0\u4e8c\u53c9\u6811\u3002<\/p>\n<p>\u8fd9\u79cd\u5bf9\u6811\u7ed3\u6784\u4e0a\u7684\u7ea6\u675f\u6709\u4e00\u5b9a\u7684\u6b63\u5219\u4f5c\u7528\u3002\u66f4\u4e3a\u91cd\u8981\u7684\u662f\uff0c\u5b83\u53ef\u4ee5\u8ba9CatBoost\u6a21\u578b\u7684\u63a8\u65ad\u8fc7\u7a0b\u6781\u5feb\u3002<\/p>\n<p>\u5bf9\u4e8eCatBoost\u7684tree\u7684\u9884\u6d4b\u8fc7\u7a0b\u6765\u8bf4\uff0c\u6bcf\u4e2a\u7279\u5f81\u7684\u5206\u88c2\u90fd\u662f\u72ec\u7acb\u7684\uff0c\u4e0d\u5206\u5148\u540e\u987a\u5e8f\uff0c\u591a\u4e2a\u6837\u672c\u53ef\u4ee5\u4e00\u8d77\u9884\u6d4b\u3002<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/lingbo.online\/wp-content\/uploads\/2023\/08\/duicheng.png\" alt=\"\" \/><\/p>\n<h4>\u57fa\u4e8eGPU\u7684\u52a0\u901f\u8bad\u7ec3<\/h4>\n<ul>\n<li><strong>\u5bc6\u96c6\u7684\u6570\u503c\u7279\u5f81\uff1a<\/strong>\u5bf9\u4e8e\u4efb\u4f55GBDT\u7b97\u6cd5\u800c\u8a00\uff0c\u6700\u5927\u7684\u96be\u70b9\u4e4b\u4e00\u5c31\u662f\u641c\u7d22\u6700\u4f73\u5206\u5272\u3002\u5c24\u5176\u662f\u5bf9\u4e8e\u5bc6\u96c6\u7684\u6570\u503c\u7279\u5f81\u6570\u636e\u96c6\u6765\u8bf4\uff0c\u8be5\u6b65\u9aa4\u662f\u5efa\u7acb\u51b3\u7b56\u6811\u65f6\u7684\u4e3b\u8981\u8ba1\u7b97\u8d1f\u62c5\u3002CatBoost\u4f7f\u7528oblivious\u51b3\u7b56\u6811\u4f5c\u4e3a\u57fa\u6a21\u578b\uff0c\u5e76\u5c06\u7279\u5f81\u79bb\u6563\u5316\u5230\u56fa\u5b9a\u6570\u91cf\u7684\u7bb1\u5b50\u4e2d\u4ee5\u51cf\u5c11\u5185\u5b58\u4f7f\u7528\u3002\u5c31GPU\u5185\u5b58\u4f7f\u7528\u800c\u8a00\uff0cCatBoost\u81f3\u5c11\u4e0eLightGBM\u4e00\u6837\u6709\u6548\u3002\u4e3b\u8981\u6539\u8fdb\u4e4b\u5904\u5c31\u662f\u5229\u7528\u4e86\u4e00\u79cd\u4e0d\u4f9d\u8d56\u4e8e\u539f\u5b50\u64cd\u4f5c\u7684\u76f4\u65b9\u56fe\u8ba1\u7b97\u65b9\u6cd5\u3002<\/li>\n<li><strong>\u7c7b\u522b\u578b\u7279\u5f81\uff1a<\/strong>CatBoost\u5b9e\u73b0\u591a\u79cd\u5904\u7406\u7c7b\u522b\u578b\u7279\u5f81\u7684\u65b9\u6cd5\uff0c\u5e76\u4f7f\u7528\u5b8c\u7f8e\u54c8\u5e0c\u6765\u5b58\u50a8\u7c7b\u522b\u578b\u7279\u5f81\u7684\u503c\uff0c\u4ee5\u51cf\u5c11\u5185\u5b58\u4f7f\u7528\u3002\u7531\u4e8eGPU\u5185\u5b58\u7684\u9650\u5236\uff0c\u5728CPU RAM\u4e2d\u5b58\u50a8\u6309\u4f4d\u538b\u7f29\u7684\u5b8c\u7f8e\u54c8\u5e0c\uff0c\u4ee5\u53ca\u8981\u6c42\u7684\u6570\u636e\u6d41\u3001\u91cd\u53e0\u8ba1\u7b97\u548c\u5185\u5b58\u7b49\u64cd\u4f5c\u3002\u901a\u8fc7\u54c8\u5e0c\u6765\u5206\u7ec4\u89c2\u5bdf\u3002\u5728\u6bcf\u4e2a\u7ec4\u4e2d\uff0c\u6211\u4eec\u9700\u8981\u8ba1\u7b97\u4e00\u4e9b\u7edf\u8ba1\u91cf\u7684\u524d\u7f00\u548c\u3002\u8be5\u7edf\u8ba1\u91cf\u7684\u8ba1\u7b97\u4f7f\u7528\u5206\u6bb5\u626b\u63cfGPU\u56fe\u5143\u5b9e\u73b0\u3002<\/li>\n<li><strong>\u591aGPU\u652f\u6301\uff1a<\/strong>CatBoost\u4e2d\u7684GPU\u5b9e\u73b0\u53ef\u652f\u6301\u591a\u4e2aGPU\u3002\u5206\u5e03\u5f0f\u6811\u5b66\u4e60\u53ef\u4ee5\u901a\u8fc7\u6570\u636e\u6216\u7279\u5f81\u8fdb\u884c\u5e76\u884c\u5316\u3002CatBoost\u91c7\u7528\u591a\u4e2a\u5b66\u4e60\u6570\u636e\u96c6\u6392\u5217\u7684\u8ba1\u7b97\u65b9\u6848\uff0c\u5728\u8bad\u7ec3\u671f\u95f4\u8ba1\u7b97\u7c7b\u522b\u578b\u7279\u5f81\u7684\u7edf\u8ba1\u6570\u636e\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u6982\u8ff0 CatBoost\u662f\u4fc4\u7f57\u65af\u7684\u641c\u7d22\u5de8\u5934Yandex\u57282017\u5e74\u5f00\u6e90\u7684\u673a\u5668\u5b66\u4e60\u5e93\uff0c\u662fBoosting\u65cf\u7b97\u6cd5\u7684\u4e00\u79cd\u3002CatBoost &#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"emotion":"","emotion_color":"","title_style":"","license":"","footnotes":""},"categories":[22,2],"tags":[36,37,38,4],"class_list":["post-305","post","type-post","status-publish","format-standard","hentry","category-data_mining","category-algorithm_learning","tag-catboost","tag-gbdt","tag-38","tag-4"],"_links":{"self":[{"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/posts\/305","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/comments?post=305"}],"version-history":[{"count":7,"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/posts\/305\/revisions"}],"predecessor-version":[{"id":317,"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/posts\/305\/revisions\/317"}],"wp:attachment":[{"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/media?parent=305"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/categories?post=305"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/lingbo.online\/index.php\/wp-json\/wp\/v2\/tags?post=305"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}