2

我是使用 TF Slim 进行培训的新手：

slim.learning.train(
    train_op,
    FLAGS.log_dir,
    log_every_n_steps=1,
    number_of_steps=FLAGS.max_steps,
    save_summaries_secs=1)

我的日志看起来像：

...
INFO:tensorflow:global step 93: loss = 0.0425 (0.25 sec/step)
INFO:tensorflow:global step 94: loss = 0.0460 (0.39 sec/step)
INFO:tensorflow:global step 95: loss = 0.0416 (0.25 sec/step)
INFO:tensorflow:global step 96: loss = 0.0480 (0.25 sec/step)
INFO:tensorflow:global step 97: loss = 0.0424 (0.27 sec/step)
INFO:tensorflow:global step 98: loss = 0.0487 (0.30 sec/step)
INFO:tensorflow:global step 99: loss = 0.0481 (0.26 sec/step)
INFO:tensorflow:global step 100: loss = 0.0390 (0.25 sec/step)

我的总结如下：

...
ScalarEvent(wall_time=1496698866.417926, step=93L, value=0.04702041298151016)
ScalarEvent(wall_time=1496698867.504077, step=97L, value=0.0489736944437027)
ScalarEvent(wall_time=1496698868.423048, step=100L, value=0.04433770477771759)
ScalarEvent(wall_time=1496698869.449098, step=100L, value=0.035440053790807724)

为什么同一时间步长的数字不同？如何让我的摘要与我的日志具有相同的数字？

4

0 回答 0