{"id":909,"date":"2021-08-05T07:43:24","date_gmt":"2021-08-04T22:43:24","guid":{"rendered":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/?p=909"},"modified":"2022-02-25T10:47:07","modified_gmt":"2022-02-25T01:47:07","slug":"improve-the-performance-of-xgboost-and-lightgbm-inference","status":"publish","type":"post","link":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/2021\/08\/05\/improve-the-performance-of-xgboost-and-lightgbm-inference\/","title":{"rendered":"XGBoost \u307e\u305f\u306f LightGBM \u30e2\u30c7\u30eb\u306e\u4e88\u6e2c\u901f\u5ea6\u3092\u6700\u5927 36 \u500d\u307e\u3067\u5411\u4e0a\u3055\u305b\u308b\u65b9\u6cd5"},"content":{"rendered":"

\u6c7a\u5b9a\u6728\u306b\u5bfe\u3059\u308b\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u306f\u3001\u5206\u985e\u3084\u56de\u5e30\u306e\u305f\u3081\u306e\u6700\u3082\u6b63\u78ba\u3067\u52b9\u7387\u7684\u306a\u30de\u30b7\u30f3\u30e9\u30fc\u30cb\u30f3\u30b0\u30fb\u30a2\u30eb\u30b4\u30ea\u30ba\u30e0\u306e 1 \u3064\u3067\u3059\u3002\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u306b\u306f\u591a\u304f\u306e\u5b9f\u88c5\u304c\u3042\u308a\u307e\u3059\u304c\u3001\u6700\u3082\u4eba\u6c17\u304c\u3042\u308b\u306e\u306f XGBoost<\/a> (\u82f1\u8a9e) \u304a\u3088\u3073 LightGBM<\/a> (\u82f1\u8a9e) \u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3059\u3002\u3053\u306e\u8a18\u4e8b\u3067\u306f\u3001\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneAPI \u30c7\u30fc\u30bf\u30fb\u30a2\u30ca\u30ea\u30c6\u30a3\u30af\u30b9\u30fb\u30e9\u30a4\u30d6\u30e9\u30ea\u30fc (\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL)\u00a0<\/a>\u3092\u4f7f\u7528\u3057\u3066\u3001XGBoost \u307e\u305f\u306f LightGBM \u30e2\u30c7\u30eb\u306e\u4e88\u6e2c\u901f\u5ea6\u3092\u6700\u5927 36 \u500d\u307e\u3067\u5411\u4e0a\u3055\u305b\u308b\u65b9\u6cd5\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n

\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0<\/h2>\n

XGBoost \u3084 LightGBM \u306e\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u306f\u3001\u73fe\u5b9f\u4e16\u754c\u306e\u3055\u307e\u3056\u307e\u306a\u554f\u984c\u3092\u89e3\u6c7a\u3059\u308b\u305f\u3081\u3001\u307e\u305f\u3001\u7814\u7a76\u3092\u884c\u3046\u305f\u3081\u3001\u305d\u3057\u3066 Kaggle*<\/a> (\u82f1\u8a9e) \u30b3\u30f3\u30c6\u30b9\u30c8\u306b\u53c2\u52a0\u3059\u308b\u305f\u3081\u306b\u3001\u591a\u304f\u306e\u4eba\u304c\u4f7f\u7528\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306f\u512a\u308c\u305f\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u767a\u63ee\u3057\u307e\u3059\u304c\u3001\u305d\u306e\u4e88\u6e2c\u901f\u5ea6\u306f\u307e\u3060\u6539\u5584\u306e\u4f59\u5730\u304c\u3042\u308a\u307e\u3059\u3002\u4e88\u6e2c\u306f\u3001\u30de\u30b7\u30f3\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u30ef\u30fc\u30af\u30d5\u30ed\u30fc\u306e\u4e2d\u3067\u6700\u3082\u91cd\u8981\u306a\u6bb5\u968e\u3067\u3042\u308b\u3053\u3068\u3092\u8003\u3048\u308b\u3068\u3001\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u5411\u4e0a\u306f\u6709\u76ca\u306a\u3082\u306e\u3067\u3059\u3002<\/p>\n

\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306f\u7af6\u5408\u4ed6\u793e\u306b\u6bd4\u3079\u3066\u6570\u500d\u306e\u901f\u5ea6\u3067\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u63a8\u8ad6\u3092\u884c\u3044\u307e\u3059 (\u9ad8\u901f\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u6728\u63a8\u8ad6<\/a> (\u82f1\u8a9e)) \u3053\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u4e0a\u306e\u30e1\u30ea\u30c3\u30c8\u304c\u3001XGBoost \u3068 LightGBM \u3067\u3082\u5229\u7528\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u308a\u307e\u3057\u305f\u3002<\/p>\n

\u30e2\u30c7\u30eb\u30fb\u30b3\u30f3\u30d0\u30fc\u30bf\u30fc<\/h2>\n

\u3059\u3079\u3066\u306e\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u306e\u5b9f\u88c5\u306f\u3001\u540c\u69d8\u306e\u64cd\u4f5c\u3092\u884c\u3046\u305f\u3081\u3001\u30c7\u30fc\u30bf\u306e\u4fdd\u5b58\u65b9\u6cd5\u3082\u540c\u69d8\u3067\u3059\u3002\u7406\u8ad6\u7684\u306b\u306f\u3001\u3053\u308c\u306b\u3088\u308a\u3001\u5b66\u7fd2\u3057\u305f\u30e2\u30c7\u30eb\u3092\u3042\u308b\u30de\u30b7\u30f3\u30e9\u30fc\u30cb\u30f3\u30b0\u30fb\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u304b\u3089\u5225\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u5909\u63db\u3059\u308b\u3053\u3068\u304c\u5bb9\u6613\u306b\u306a\u308a\u307e\u3059\u3002\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306e\u30e2\u30c7\u30eb\u30fb\u30b3\u30f3\u30d0\u30fc\u30bf\u30fc\u306f\u3001XGBoost \u3084 LightGBM \u3067\u5b66\u7fd2\u3057\u305f\u30e2\u30c7\u30eb\u3092\u3001\u305f\u3063\u305f 1 \u884c\u306e\u30b3\u30fc\u30c9\u3067\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306b\u8ee2\u9001\u3067\u304d\u308b\u3088\u3046\u306b\u8a2d\u8a08\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u307b\u304b\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u30e2\u30c7\u30eb\u30fb\u30b3\u30f3\u30d0\u30fc\u30bf\u30fc\u3082\u8fd1\u65e5\u4e2d\u306b\u5229\u7528\u53ef\u80fd\u306b\u306a\u308b\u4e88\u5b9a\u3067\u3059\u3002<\/p>\n

\u3053\u306e\u4f8b\u3067\u306f\u3001XGBoost \u3068 LightGBM \u30e2\u30c7\u30eb\u3092\u30a4\u30f3\u30c6\u30eb\u00ae oneDAL \u306b\u5909\u63db\u3059\u308b\u65b9\u6cd5\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n

\u307e\u305a\u3001Python* 3.6 \u4ee5\u4e0a\u306b\u5bfe\u5fdc\u3057\u305f\u6700\u65b0\u30d0\u30fc\u30b8\u30e7\u30f3\u306e daal4py<\/a> \u3092\u5165\u624b\u3057\u307e\u3059\u3002<\/p>\n

conda install -c conda-forge daal4py'>=2020.3'\r\n<\/pre>\n
XGBoost \u30e2\u30c7\u30eb\u3092\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306b\u5909\u63db:<\/strong><\/p>\n
# Train an XGBoost model\r\nimport xgboost as xgb\r\nclf = xgb.XGBClassifier(**params)\r\nxgb_model = clf.fit(X_train, y_train)\r\n\r\n# Convert the XGBoost model to a oneDAL model\r\nimport daal4py as d4p\r\ndaal_model = d4p.get_gbt_model_from_xgboost(xgb_model.get_booster())\r\n\r\n# Make a faster prediction with oneDAL\r\ndeel_prediction = d4p.gbt_classification_prediction(nClasses=n_classes)\r\n.compute(X_test, daal_model).prediction<\/pre>\n
LightGBM \u30e2\u30c7\u30eb\u3092\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306b\u5909\u63db:<\/strong><\/p>\n
# Train a LightGBM model\r\nimport lightgbm as lgb\r\nlgb_model = lgb.train(params, lgb.Dataset(X_train, y_train))\r\n\r\n# Convert the LightGBM model to a oneDAL model\r\nimport deel4py as d4p\r\ndaal_model = d4p.get_gbt_model_from_lightgbm(lgb_model)\r\n\r\n# Make a faster prediction with oneDAL\r\ndaal_prediction = d4p.gbt_regression_predition().compute(X_test,\r\ndaal_model).predition<\/pre>\n
\u6ce8) \u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3084\u4e88\u6e2c\u306e\u969b\u306b\u6b20\u640d\u5024 (NaN) \u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u306b\u4e00\u6642\u7684\u306a\u5236\u9650\u304c\u3042\u308a\u307e\u3059\u3002\u30c7\u30fc\u30bf\u306b\u6b20\u640d\u5024\u304c\u3042\u308b\u3068\u3001\u63a8\u8ad6\u306e\u8cea\u304c\u4f4e\u4e0b\u3059\u308b\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n
\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u304b\u3089\u30e2\u30c7\u30eb\u3092\u4fdd\u5b58\u30fb\u8aad\u307f\u8fbc\u307f\u3059\u308b\u65b9\u6cd5\u306e\u4e00\u4f8b:<\/strong><\/p>\n
# Model from XGBoost\r\ndaal_model = d4p.get_gbt_model_from_xgboost(xgb_model)\r\n\r\nimport pickle\r\n\r\n# Save model to a file \r\nwith open('model.pkl','wb') as out:\r\n   pickle.dump(daal_model, out)\r\n\r\n# Load model from a file \r\nwith open('model.pkl','rb') as inp:\r\n   model = pickle.load(inp)\r\n\r\n# Make predictions\r\ndaal_prediction =\r\nd4p.gbt_regression_prediction().compute(X_test, model)\r\n<\/pre>\n
\u30c7\u30d5\u30a9\u30eb\u30c8\u3067\u306f\u3001\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306f\u4e88\u6e2c\u3055\u308c\u305f\u8981\u7d20\u306e\u30e9\u30d9\u30eb\u306e\u307f\u3092\u8fd4\u3057\u307e\u3059\u3002\u78ba\u7387\u3082\u5fc5\u8981\u306a\u5834\u5408\u306f\u3001\u660e\u793a\u7684\u306b\u8981\u6c42\u3059\u308b\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n
# List all results that you need by placing '|' between them:\r\npredict_algo = d4p.gbt_classfication_prediction(nClasses=n_classes,\r\n    resultsToEvaluate=\"computeClassLabels|computeClassProbabilities\")\r\ndaal_prediction = predict_algo.compute(X_test, model)\r\n\r\n# Get probabilities:\r\nprobabilities = daal_prediction.probabilities\r\n\r\n# Get labels:\r\nlabels = daal_prediction.prediction<\/pre>\n
\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u6bd4\u8f03<\/h2>\n
\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u304c XGBoost \u3084 LightGBM \u3088\u308a\u3082\u512a\u308c\u3066\u3044\u308b\u3053\u3068\u306f\u3001\u4ee5\u4e0b\u306e\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u4f7f\u3063\u3066\u5b9f\u8a3c\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n
\u30fbMortgage<\/a> (45 \u500b\u306e\u7279\u5fb4\u3001900 \u4e07\u500b\u306e\u89b3\u6e2c\u5024)<\/p>\n
\u30fbAirline<\/a>\u00a0(691 \u500b\u306e\u7279\u5fb4\u3001\u30ef\u30f3\u30b7\u30e7\u30c3\u30c8\u30fb\u30a8\u30f3\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u3001100 \u4e07\u500b\u306e\u89b3\u6e2c\u5024)<\/p>\n
\u30fbHiggs<\/a> (28 \u500b\u306e\u7279\u5fb4\u3001100 \u4e07\u500b\u306e\u89b3\u6e2c\u5024)<\/p>\n
\u30fbMSRank<\/a> (136 \u500b\u306e\u7279\u5fb4\u3001300 \u4e07\u500b\u306e\u89b3\u6e2c\u5024)<\/p>\n
\u30e2\u30c7\u30eb\u306f\u3001XGBoost \u3068 LightGBM \u3067\u5b66\u7fd2\u3057\u305f\u5f8c\u3001daal4py \u306b\u5909\u63db\u3057\u307e\u3057\u305f\u3002\u30aa\u30ea\u30b8\u30ca\u30eb\u306e XGBoost\/LightGBM \u30e2\u30c7\u30eb\u3068 daal4py \u3067\u5909\u63db\u3055\u308c\u305f\u30e2\u30c7\u30eb\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u6bd4\u8f03\u3059\u308b\u305f\u3081\u306b\u3001\u30aa\u30ea\u30b8\u30ca\u30eb\u306e\u30e2\u30c7\u30eb\u3068\u5909\u63db\u3055\u308c\u305f\u30e2\u30c7\u30eb\u306e\u4e21\u65b9\u306e\u4e88\u6e2c\u6642\u9593\u3092\u6e2c\u5b9a\u3057\u307e\u3057\u305f\u3002\u56f3 1 \u306b\u3088\u308b\u3068\u3001daal4py \u306f XGBoost \u3088\u308a\u3082\u6700\u5927 36 \u500d (\u5e73\u5747 24 \u500d)\u3001LightGBM \u3088\u308a\u3082\u6700\u5927 15.5 \u500d (\u5e73\u5747 14.5 \u500d) \u9ad8\u901f\u3067\u3042\u308b\u3053\u3068\u304c\u308f\u304b\u308a\u307e\u3059\u3002\u4e88\u6e2c\u306e\u8cea\u306f\u5909\u308f\u308a\u307e\u305b\u3093 (\u56de\u5e30\u3067\u306f\u5e73\u5747\u4e8c\u4e57\u8aa4\u5dee\u3001\u5206\u985e\u3067\u306f\u7cbe\u5ea6\u3068\u30ed\u30b8\u30b9\u30c6\u30a3\u30c3\u30af\u640d\u5931\u3067\u6e2c\u5b9a\uff09\u3002<\/p>\n
\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306f\u3001\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> \u30a2\u30c9\u30d0\u30f3\u30b9\u30c8\u30fb\u30d9\u30af\u30c8\u30eb\u30fb\u30a8\u30af\u30b9\u30c6\u30f3\u30b7\u30e7\u30f3 512<\/a> (\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> AVX-512) \u547d\u4ee4\u30bb\u30c3\u30c8\u3092\u4f7f\u7528\u3057\u3066\u3001\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> Xeon\u00ae<\/sup> \u30d7\u30ed\u30bb\u30c3\u30b5\u30fc<\/a>\u306b\u304a\u3051\u308b\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u6700\u5927\u9650\u306b\u5f15\u304d\u51fa\u3057\u307e\u3059\u3002\u6bd4\u8f03\u3084\u30e9\u30f3\u30c0\u30e0\u30fb\u30e1\u30e2\u30ea\u30fc\u30fb\u30a2\u30af\u30bb\u30b9\u306a\u3069\u3001\u6700\u3082\u4e00\u822c\u7684\u306b\u4f7f\u7528\u3055\u308c\u308b\u63a8\u8ad6\u64cd\u4f5c\u306f\u3001\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> AVX-512 \u306e vpgatherd{d,q} \u547d\u4ee4\u3068 vcmpp{s,d} \u547d\u4ee4\u3092\u4f7f\u7528\u3057\u3066\u52b9\u679c\u7684\u306b\u5b9f\u88c5\u3067\u304d\u307e\u3059\u3002\u307e\u305f\u3001\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306f\u30b9\u30c8\u30ec\u30fc\u30b8\u306e\u52b9\u7387\u3084\u30e1\u30e2\u30ea\u30fc\u306e\u5e2f\u57df\u5e45\u306b\u3082\u4f9d\u5b58\u3057\u307e\u3059\u3002\u6728\u69cb\u9020\u306e\u5834\u5408\u3001\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u306f\u30e1\u30e2\u30ea\u30fc\u5185\u306e\u30c7\u30fc\u30bf\u3092\u30b9\u30de\u30fc\u30c8\u30ed\u30c3\u30af\u3059\u308b\u3053\u3068\u3067\u3001\u4e00\u6642\u7684\u306a\u30ad\u30e3\u30c3\u30b7\u30e5\u30fb\u30ed\u30fc\u30ab\u30ea\u30bc\u30fc\u30b7\u30e7\u30f3 (\u6728\u306e\u30b5\u30d6\u30bb\u30c3\u30c8\u3068\u89b3\u6e2c\u5024\u306e\u30d6\u30ed\u30c3\u30af\u304c L1 \u30c7\u30fc\u30bf\u30ad\u30e3\u30c3\u30b7\u30e5\u306b\u683c\u7d0d\u3055\u308c\u3066\u3044\u308b\u72b6\u614b) \u3092\u5b9f\u73fe\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u306e\u52d5\u4f5c\u306b\u3088\u308a\u3001\u307b\u3068\u3093\u3069\u306e\u30e1\u30e2\u30ea\u30fc\u30a2\u30af\u30bb\u30b9\u304c\u3001\u30e1\u30e2\u30ea\u30fc\u5e2f\u57df\u5e45\u304c\u6700\u3082\u9ad8\u3044 L1 \u30ec\u30d9\u30eb\u3067\u76f4\u3061\u306b\u6e80\u305f\u3055\u308c\u307e\u3059\u3002<\/p>\n
\"\"<\/p>\n
\u56f3 1. \u30aa\u30ea\u30b8\u30ca\u30eb\u306e XGBoost\/LightGBM \u30e2\u30c7\u30eb\u3068 daal4py \u3067\u5909\u63db\u3055\u308c\u305f\u30e2\u30c7\u30eb\u306e\u63a8\u8ad6\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u6bd4\u8f03<\/p>\n
\u307e\u3068\u3081<\/h2>\n
\u591a\u304f\u306e\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u3067\u306f\u3001XGBoost \u3084 LightGBM \u3092\u4f7f\u3063\u3066\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u304c\u3001\u30e2\u30c7\u30eb\u30fb\u30b3\u30f3\u30d0\u30fc\u30bf\u30fc\u3092\u4f7f\u3046\u3053\u3068\u3067\u3001\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneDAL \u3092\u4f7f\u3063\u305f\u63a8\u8ad6\u3092\u7c21\u5358\u306b\u9ad8\u901f\u5316\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002\u30e2\u30c7\u30eb\u30fb\u30b3\u30f3\u30d0\u30fc\u30bf\u30fc\u3092\u4f7f\u3046\u3068\u3001XGBoost \u3068 LightGBM \u306e\u30e6\u30fc\u30b6\u30fc\u306f\u4ee5\u4e0b\u306e\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002<\/p>\n
\u30fb\u65e2\u5b58\u306e\u30e2\u30c7\u30eb\u306e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30fb\u30b3\u30fc\u30c9\u3092\u5909\u66f4\u305b\u305a\u306b\u4f7f\u7528<\/p>\n
\u30fb\u30b3\u30fc\u30c9\u306e\u5909\u66f4\u3092\u6700\u5c0f\u9650\u306b\u6291\u3048\u3001\u54c1\u8cea\u3092\u640d\u306a\u3046\u3053\u3068\u306a\u304f\u3001\u6700\u5927 36 \u500d\u306e\u63a8\u8ad6\u901f\u5ea6\u3092\u5b9f\u73fe<\/p>\n
\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u3068\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u306e\u69cb\u6210<\/h2>\n
\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> Xeon\u00ae<\/sup> Platinum 8275CL \u30d7\u30ed\u30bb\u30c3\u30b5\u30fc (\u7b2c 2 \u4e16\u4ee3\u30a4\u30f3\u30c6\u30eb\u00ae Xeon\u00ae \u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u30fb\u30d7\u30ed\u30bb\u30c3\u30b5\u30fc): 2 \u30bd\u30b1\u30c3\u30c8\u3001\u30bd\u30b1\u30c3\u30c8\u3042\u305f\u308a 24 \u30b3\u30a2<\/p>\n
\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> \u30cf\u30a4\u30d1\u30fc\u30b9\u30ec\u30c3\u30c7\u30a3\u30f3\u30b0\u30fb\u30c6\u30af\u30ce\u30ed\u30b8\u30fc: \u6709\u52b9<\/p>\n
\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> \u30bf\u30fc\u30dc\u30fb\u30d6\u30fc\u30b9\u30c8\u30fb\u30c6\u30af\u30ce\u30ed\u30b8\u30fc: \u6709\u52b9<\/p>\n
\u30aa\u30da\u30ec\u30fc\u30c6\u30a3\u30f3\u30b0\u30fb\u30b7\u30b9\u30c6\u30e0: Ubuntu* 18.04.4 LTS (Bionic Beaver)\u3001\u5408\u8a08\u30e1\u30e2\u30ea\u30fc 192GB (12 \u30b9\u30ed\u30c3\u30c8\/16GB\/2933MHz)<\/p>\n
\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2: XGBoost 1.2.1\u3001LightGBM 3.0.0\u3001daal4py \u30d0\u30fc\u30b8\u30e7\u30f3 2020 Update 3\u3001Python* 3.7.9\u3001NumPy 1.19.2\u3001pandas 1.1.3\u3001scikit-learn 0.23.2<\/p>\n
\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30fb\u30d1\u30e9\u30e1\u30fc\u30bf\u30fc: XGBoost<\/a> (\u82f1\u8a9e) \u304a\u3088\u3073 LightGBM<\/a> (\u82f1\u8a9e)<\/p>\n
\n
\u30a4\u30f3\u30c6\u30eb\u00ae<\/sup> oneAPI \u306b\u95a2\u3059\u308b\u8a73\u7d30\u306f\u3053\u3061\u3089<\/a>\u304b\u3089\u3054\u53c2\u7167\u304f\u3060\u3055\u3044\u3002<\/p>\n
\n
\u53c2\u7167\u8a18\u4e8b:
\nImprove the Performance of XGBoost and LightGBM Inference<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"
\u6c7a\u5b9a\u6728\u306b\u5bfe\u3059\u308b\u52fe\u914d\u30d6\u30fc\u30b9\u30c6\u30a3\u30f3\u30b0\u306f\u3001\u5206\u985e\u3084\u56de\u5e30\u306e\u305f […]<\/a><\/p>\n<\/div>","protected":false},"author":1,"featured_media":913,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[27,28,21,25],"class_list":["post-909","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-1","tag-gboost","tag-lightgbm","tag-onedal","tag-25"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts\/909","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/comments?post=909"}],"version-history":[{"count":13,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts\/909\/revisions"}],"predecessor-version":[{"id":1913,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/posts\/909\/revisions\/1913"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/media\/913"}],"wp:attachment":[{"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/media?parent=909"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/categories?post=909"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.xlsoft.com\/jp\/blog\/intel\/wp-json\/wp\/v2\/tags?post=909"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}