在TensorBoard上使用Estimator测试错误和准确性图

Question

我对TensorFlow相对陌生，并且在TensorBoard和Estimator API中绝对是新事物。我想从给出的源代码中训练Tensorflow的经过稍微修改的Resnet模型，作为正式模型，它使用Estimator。

我需要测试误差和准确性图。精度图只显示了一个点，根本没有测试错误。我还需要x轴上的历元数，而不是步骤数。使用低级Tensorflow可以轻松实现这一点，但是我需要使用给出的模型。

我在resnet_fn中创建learning_rate，cross_entropy和train_accuracy张量，如下所示。我还在resnet_fn中添加了SummarySaverHook。这也没有帮助。

 def resnet_model_fn(...):

    ...

    tf.identity(learning_rate, name='learning_rate')
    tf.summary.scalar('learning_rate', learning_rate)

    ....

    summary_hook = tf.train.SummarySaverHook(
        flags.epochs_per_eval,
        output_dir=FLAGS.model_dir,
        summary_op=tf.summary.merge_all())

      return tf.estimator.EstimatorSpec(
          mode=mode,
          predictions=predictions,
          loss=loss,
          train_op=train_op,
          eval_metric_ops=metrics,
          training_hooks=[summary_hook],
          evaluation_hooks=[summary_hook])

这是resnet_main（）。我可以在终端上看到那些张量“ eval_cross_entropy”等，但是它们在TensorBoard中根本没有显示。我也正在分享TensorBoard的屏幕截图。

def resnet_main(flags, model_function, input_function):

  # Using the Winograd non-fused algorithms provides a small performance boost.
  os.environ['TF_ENABLE_WINOGRAD_NONFUSED'] = '1'

  # Set up a RunConfig to only save checkpoints once per training cycle.
  run_config = tf.estimator.RunConfig().replace(save_checkpoints_secs=1e9, save_summary_steps=flags.epochs_per_eval)
  classifier = tf.estimator.Estimator(
      model_fn=model_function, model_dir=flags.model_dir, config=run_config,
      params={
          'resnet_size': flags.resnet_size,
          'data_format': flags.data_format,
          'batch_size': flags.batch_size,
      })

  for _ in range(flags.train_epochs // flags.epochs_per_eval):
    tensors_to_log = {
        'learning_rate': 'learning_rate',
        'cross_entropy': 'cross_entropy',
        'train_accuracy': 'train_accuracy'
    }

    logging_hook = tf.train.LoggingTensorHook(
        tensors=tensors_to_log, every_n_iter=flags.epochs_per_eval)

    print('Starting a training cycle.')

    def input_fn_train():
      return input_function(True, flags.data_dir, flags.batch_size,
                            flags.epochs_per_eval, flags.num_parallel_calls)

    classifier.train(input_fn=input_fn_train, hooks=[logging_hook])

    tensors_to_log_eval = {
        'eval_cross_entropy': 'cross_entropy',
        'eval_train_accuracy': 'train_accuracy'
    }

    logging_hook_eval = tf.train.LoggingTensorHook(
        tensors=tensors_to_log_eval, every_n_iter=flags.epochs_per_eval)

    print('Starting to evaluate.')
    # Evaluate the model and print results
    def input_fn_eval():
      return input_function(False, flags.data_dir, flags.batch_size,
                            1, flags.num_parallel_calls)

    eval_results = classifier.evaluate(input_fn=input_fn_eval, hooks=[logging_hook_eval])

    tensors_to_log_pred = {
        'pred_cross_entropy': 'cross_entropy'
    }

    logging_hook_pred = tf.train.LoggingTensorHook(
        tensors=tensors_to_log_pred, every_n_iter=flags.epochs_per_eval)

    print('Starting to predict.')

    def input_fn_pred():
      return input_function(False, flags.data_dir, flags.batch_size,
                            1, flags.num_parallel_calls)

    pred_results = classifier.predict(input_fn=input_fn_pred, hooks=[logging_hook_pred])

    return eval_results, pred_results

如何获得带有x轴历元数的测试误差和准确性图？