如何在tensorflow r0.9(skflow)中训练DNNC分类器时打印进度？

Question

我无法得到DNNClassifier在训练时打印进度,即损失和验证分数.据我所知,可以使用继承自BaseEstimator的config参数打印丢失,但是当我传递RunConfig对象时,分类器没有打印任何内容.

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,它似乎只关心它是否大于1,这与文档不匹配:

verbose:控制详细程度,可能的值:0:算法和调试信息被静音.1:培训师打印进度.2:打印日志设备放置.

至于验证分数,我认为使用monitors.ValidationMonitor会很好,但是当尝试它时,分类器不会打印任何东西,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,但我找不到任何监视器.

Answer 1

在fit函数之前添加这些以显示进度:

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

样品:

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