本文主要是介绍MNIST2_LGB_XGB训练预测，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！
针对MNIST数据集进行XGB\LGB模型训练和预测
部分脚本如下： 完整脚本见笔者github
lgb_param = {'boosting': 'gbdt','num_iterations': 145,'num_threads' : 8, 'verbosity': 0,'learning_rate': 0.2,'max_depth' : 10,'num_leaves' : 8,'subsample' : 0.75,'subsample_freq': 5,'colsample_bytree' : 1,'reg_alpha': 1.5,'reg_lambda': 0.75,'objective': 'multiclass','num_class': 10,'metric': 'multi_logloss','early_stopping': 25# 'device': 'gpu',# 'gpu_platform_id': 0,# 'gpu_device_id': 0
}xgb_param = {'booster': 'gbtree','tree_method':'gpu_hist','num_rounds': 160,'nthread' : 8, 'silent' : 1,'learning_rate': 0.2,'max_depth' : 10,'num_leaves' : 8,'subsample' : 0.75,'colsample_bytree' : 1,'reg_alpha': 1.5,'reg_lambda': 0.75,'objective': 'multi:softprob','num_class': 10,'metric': 'mlogloss','early_stopping': 25
}@clock
def lgb_xgb_train(model, param, tr, te ):if model.__name__ == 'lightgbm':trdt = model.Dataset(data=tr.iloc[:, :-1].values, label=tr.iloc[:, -1].values)tedt = model.Dataset(data=te.iloc[:, :-1].values, label=te.iloc[:, -1].values)clf_model = model.train(param, trdt, valid_sets=[trdt, tedt] ,verbose_eval = 20)pred = np.argmax(clf_model.predict(te.iloc[:, :-1].values, num_iteration=clf_model.best_iteration ), axis=1)else:trdt = model.DMatrix(data=tr.iloc[:, :-1].values, label=tr.iloc[:, -1].values)tedt = model.DMatrix(data=te.iloc[:, :-1].values, label=te.iloc[:, -1].values)clf_model = model.train(param, trdt, evals=[(trdt, 'train'), (tedt, 'test')], verbose_eval = 20)pred = np.argmax(clf_model.predict(tedt, ntree_limit=-1), axis=1)y_te = te.iloc[:, -1].valuesacc_ = sum(pred == y_te)/len(y_te) * 100return f'model: {model.__name__}, acc: {acc_:.2f}'if __name__ == '__main__':mnistdf = get_ministdata()te_index = mnistdf.sample(frac=0.8).index.tolist()mnist_te = mnistdf.loc[te_index, :]mnist_tr = mnistdf.loc[~mnistdf.index.isin(te_index), :]print('train xgb ...')resxgb = lgb_xgb_train(xgb, xgb_param, mnist_tr, mnist_te)print('train lgb ...')reslgb = lgb_xgb_train(lgb, lgb_param, mnist_tr, mnist_te)- 结果如下
train xgb ...
[0] train-merror:0.078143 test-merror:0.144911
[9] train-merror:0.013643 test-merror:0.070464
lgb_xgb_train, take_time:37.27306s >> model: xgboost, acc: 92.95
train lgb ...
Training until validation scores don't improve for 25 rounds.
[20] training's multi_logloss: 0.360737 valid_1's multi_logloss: 0.419623
[40] training's multi_logloss: 0.178201 valid_1's multi_logloss: 0.259657
[60] training's multi_logloss: 0.110021 valid_1's multi_logloss: 0.206787
[80] training's multi_logloss: 0.0729135 valid_1's multi_logloss: 0.180611
[100] training's multi_logloss: 0.051499 valid_1's multi_logloss: 0.16564
[120] training's multi_logloss: 0.0381409 valid_1's multi_logloss: 0.156427
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[140] training's multi_logloss: 0.0299155 valid_1's multi_logloss: 0.151268
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
Did not meet early stopping. Best iteration is:
[145] training's multi_logloss: 0.0283099 valid_1's multi_logloss: 0.150082
lgb_xgb_train, take_time:27.31041s >> model: lightgbm, acc: 95.43
这篇关于MNIST2_LGB_XGB训练预测的文章就介绍到这儿，希望我们推荐的文章对编程师们有所帮助！
