将 XGBoost 的 xgb.train 的输出保存为带有 python 日志记录的日志文件

Question

我试图通过 将xgb.trainXGBoost的输出保存为日志文件logging，但我无法记录输出。我怎样才能记录它？我试图参考现有的 Stackoverflow 问题，但这是不可能的。我希望你用一个具体的样本来展示它。

import sys
import logging

# ---------------------------------------------- #
# Some logging settings
# ---------------------------------------------- #

import xgboost as xgb

import numpy as np
from sklearn.model_selection import KFold
from sklearn.datasets import load_digits

rng = np.random.RandomState(31337)

print("Zeros and Ones from the Digits dataset: binary classification")
digits = load_digits(2)
y = digits['target']
X = digits['data']
kf = KFold(n_splits=2, shuffle=True, random_state=rng)
for train_index, test_index in kf.split(X):

    param = {'max_depth':2, 'eta':0.3, 'silent':1, 'objective':'binary:logistic' }

    dtrain = xgb.DMatrix(X[train_index], y[train_index])
    dtest = xgb.DMatrix(X[test_index], y[test_index])

    # specify validations set to watch performance
    watchlist  = [(dtest,'eval'), (dtrain,'train')]
    num_round = 2
    bst = xgb.train(param, dtrain, num_round, watchlist)

# I want to record this output.
# Zeros and Ones from the Digits dataset: binary classification
# [0]   eval-error:0.011111 train-error:0.011111
# [1]   eval-error:0.011111 train-error:0.005556
# [0]   eval-error:0.016667 train-error:0.005556
# [1]   eval-error:0.005556 train-error:0

Answer 1

xgboost 将其日志直接打印到标准输出中，您无法更改其行为。但callbacks参数xgb.train有能力记录结果与内部打印相同的时间。

以下代码是使用回调将 xgboost 日志记录到记录器的示例。 log_evaluation()返回一个从 xgboost 内部调用的回调函数，您可以将回调函数添加到callbacks

from logging import getLogger, basicConfig, INFO

import numpy as np
import xgboost as xgb
from sklearn.datasets import load_digits
from sklearn.model_selection import KFold

# Some logging settings
basicConfig(level=INFO)
logger = getLogger(__name__)


def log_evaluation(period=1, show_stdv=True):
    """Create a callback that logs evaluation result with logger.

    Parameters
    ----------
    period : int
        The period to log the evaluation results

    show_stdv : bool, optional
         Whether show stdv if provided

    Returns
    -------
    callback : function
        A callback that logs evaluation every period iterations into logger.
    """

    def _fmt_metric(value, show_stdv=True):
        """format metric string"""
        if len(value) == 2:
            return '%s:%g' % (value[0], value[1])
        elif len(value) == 3:
            if show_stdv:
                return '%s:%g+%g' % (value[0], value[1], value[2])
            else:
                return '%s:%g' % (value[0], value[1])
        else:
            raise ValueError("wrong metric value")

    def callback(env):
        if env.rank != 0 or len(env.evaluation_result_list) == 0 or period is False:
            return
        i = env.iteration
        if i % period == 0 or i + 1 == env.begin_iteration or i + 1 == env.end_iteration:
            msg = '\t'.join([_fmt_metric(x, show_stdv) for x in env.evaluation_result_list])
            logger.info('[%d]\t%s\n' % (i, msg))

    return callback


rng = np.random.RandomState(31337)

print("Zeros and Ones from the Digits dataset: binary classification")
digits = load_digits(2)
y = digits['target']
X = digits['data']
kf = KFold(n_splits=2, shuffle=True, random_state=rng)
for train_index, test_index in kf.split(X):
    param = {'max_depth': 2, 'eta': 0.3, 'silent': 1, 'objective': 'binary:logistic'}

    dtrain = xgb.DMatrix(X[train_index], y[train_index])
    dtest = xgb.DMatrix(X[test_index], y[test_index])

    # specify validations set to watch performance
    watchlist = [(dtest, 'eval'), (dtrain, 'train')]
    num_round = 2
    # add logger
    callbacks = [log_evaluation(1, True)]
    bst = xgb.train(param, dtrain, num_round, watchlist, callbacks=callbacks)

Answer 2

接受的解决方案不适用于 xgboost 版本 1.3 及更高版本。（在1.6.1上测试），由于以下原因：

在XGBoost 1.3中，为Python包设计了一个新的回调接口。

（来源：https ://xgboost.readthedocs.io/en/latest/python/callbacks.html ）

您可以通过定义自定义日志记录回调并将其作为参数传递给 xgb.train 来实现 xgboost.train 的 python 日志记录，如下所示：

import logging
logger = logging.getLogger(__name__)

import xgboost

class XGBLogging(xgboost.callback.TrainingCallback):
    """log train logs to file"""

    def __init__(self, epoch_log_interval=100):
        self.epoch_log_interval = epoch_log_interval

    def after_iteration(self, model, epoch, evals_log):
        if epoch % self.epoch_log_interval == 0:
            for data, metric in evals_log.items():
                metrics = list(metric.keys())
                metrics_str = ""
                for m_key in metrics:
                    metrics_str = metrics_str + f"{m_key}: {metric[m_key][-1]}"
                logger.info(f"Epoch: {epoch}, {data}: {metrics_str}")
        # False to indicate training should not stop.
        return False

model = xgboost.train(
                xgboost_parms, 
                dtrain=dtrain,
                evals=[(dtrain,"train"),(dvalid,"valid")]
                callbacks=[XGBLogging(epoch_log_interval=100)]
            )