Method _log_metrics

tensorflow/python/keras/callbacks.py:1707–1748

Writes metrics out as custom scalar summaries. Arguments: logs: Dict. Keys are scalar summary names, values are NumPy scalars. prefix: String. The prefix to apply to the scalar summary names. step: Int. The global step to use for TensorBoard.

(self, logs, prefix, step)

      self._is_tracing = False

  def _log_metrics(self, logs, prefix, step):
    """Writes metrics out as custom scalar summaries.

    Arguments:
        logs: Dict. Keys are scalar summary names, values are NumPy scalars.
        prefix: String. The prefix to apply to the scalar summary names.
        step: Int. The global step to use for TensorBoard.
    """
    if logs is None:
      logs = {}

    # Group metrics by the name of their associated file writer. Values
    # are lists of metrics, as (name, scalar_value) pairs.
    logs_by_writer = {
        self._train_run_name: [],
        self._validation_run_name: [],
    }
    validation_prefix = 'val_'
    for (name, value) in logs.items():
      if name in ('batch', 'size', 'num_steps'):
        # Scrub non-metric items.
        continue
      if name.startswith(validation_prefix):
        name = name[len(validation_prefix):]
        writer_name = self._validation_run_name
      else:
        writer_name = self._train_run_name
      name = prefix + name  # assign batch or epoch prefix
      logs_by_writer[writer_name].append((name, value))

    with context.eager_mode():
      with summary_ops_v2.always_record_summaries():
        for writer_name in logs_by_writer:
          these_logs = logs_by_writer[writer_name]
          if not these_logs:
            # Don't create a "validation" events file if we don't
            # actually have any validation data.
            continue
          writer = self._get_writer(writer_name)
          with writer.as_default():
            for (name, value) in these_logs:
              summary_ops_v2.scalar(name, value, step=step)

  def _log_weights(self, epoch):
    """Logs the weights of the Model to TensorBoard."""