python - 在 tensorflow r0.9 (skflow) 中训练 DNNClassifier 时如何打印进度？

Question

我无法让 DNNClassifier 在训练时打印进度，即损失和验证分数。据我所知，可以使用从 BaseEstimator 继承的配置参数打印损失，但是当我传递一个 RunConfig 对象时，分类器没有打印任何东西。

from tensorflow.contrib.learn.python.learn.estimators import run_config

config = run_config.RunConfig(verbose=1)
classifier = learn.DNNClassifier(hidden_units=[10, 20, 10],
                             n_classes=3,
                             config=config)
classifier.fit(X_train, y_train, steps=1000)

我错过了什么吗？我检查了 RunConfig 如何处理详细参数，它似乎是 that it only cares if its greater than 1 ，与文档不匹配:

verbose: Controls the verbosity, possible values: 0: the algorithm and debug information is muted. 1: trainer prints the progress. 2: log device placement is printed.

至于验证分数，我认为使用 monitors.ValidationMonitor会很好，但是当尝试它时，分类器不打印任何东西，当尝试使用 early_stopping_rounds 时也没有任何反应。我在源代码中搜索文档或一些注释，但找不到任何关于监视器的内容。

Answer 1

在 fit 函数显示进度之前添加这些:

import logging
logging.getLogger().setLevel(logging.INFO)

示例:

INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:Training steps [0,1000000)
INFO:tensorflow:Step 1: loss = 10.5043
INFO:tensorflow:training step 100, loss = 10.45380 (0.223 sec/batch).
INFO:tensorflow:Step 101: loss = 10.5623
INFO:tensorflow:training step 200, loss = 10.46701 (0.220 sec/batch).
INFO:tensorflow:Step 201: loss = 10.3885
INFO:tensorflow:training step 300, loss = 10.36501 (0.232 sec/batch).
INFO:tensorflow:Step 301: loss = 10.3441
INFO:tensorflow:training step 400, loss = 10.44571 (0.220 sec/batch).
INFO:tensorflow:Step 401: loss = 10.396
INFO:tensorflow:global_step/sec: 3.95