从原生 xgboost.train() 获取 Pandas Dataframe 结果文件答案

【问题标题】：Getting a Pandas Dataframe result file from native xgboost.train()从原生 xgboost.train() 获取 Pandas Dataframe 结果文件
【发布时间】：2023-01-17 22:12:01
【问题描述】：

可以使用参数 evals_result 从 xgboost.train() 获取结果文件，如下所示：

import xgboost as xgb
import sklearn as sk


def metric_r2(predt: np.ndarray, dtrain: xgb.DMatrix) -> Tuple[str, float]:
    """ compute R^2` (coefficient of determination) regression score function"""
    y = dtrain.get_label()
    return "R2", sk.metrics.r2_score(y, predt)


results: Dict[str, Dict[str, Union[List[float], List[Tuple[float, float]]]]] = {}
xgb_model: xgb.Booster = xgb.train(
    params, d_train, num_boost_round=100,
    evals=[(d_train, "train"), (d_test, "test")], verbose_eval=True, custom_metric=metric_r2,
    evals_result=results
)

如何将此文件转换为以下 pandas Dataframe：

   train-rmse  test-rmse  train-R2 test-R2
0     0.48253   0.480385      0.73    0.72
1     0.35953   0.357756      0.76    0.74

？

【问题讨论】：

标签： python xgboost

【解决方案1】：

我尝试了以下并且有效：

import xgboost as xgb
import pandas as pd
import sklearn as sk


def metric_r2(predt: np.ndarray, dtrain: xgb.DMatrix) -> Tuple[str, float]:
    """ compute R^2` (coefficient of determination) regression score function"""
    y = dtrain.get_label()
    return "R2", sk.metrics.r2_score(y, predt)

def convert_results_to_pandas(results: Dict[str, Dict[str, Union[List[float], List[Tuple[float, float]]]]]) -> pd.DataFrame:
    """Converts a given results dictionary into a pandas dataframe."""
    df: pd.DataFrame = pd.DataFrame()
    for dataset_name, values1 in results.items():
        for metric_name, values2 in values1.items():
            column_name = dataset_name + "-" + metric_name
            df[column_name] = values2
    return df


xgb_model: xgb.Booster = xgb.train(
    params, self.d_train, num_boost_round=self.number_of_boosting_rounds,
    evals=[(self.d_train, "train"), (self.d_test, "test")], verbose_eval=True, custom_metric=metric_r2,
    evals_result=results
)
pd_results: pd.DataFrame = convert_results_to_pandas(results)

【讨论】：