tf.contrib.checkpoint.dot_graph_from_checkpoint

View source on GitHub

Visualizes an object-based checkpoint (from tf.train.Checkpoint).

tf.contrib.checkpoint.dot_graph_from_checkpoint(save_path)

Useful for inspecting checkpoints and debugging loading issues.

Example usage from Python (requires pydot):

import tensorflow as tf
import pydot

dot_string = tf.contrib.checkpoint.dot_graph_from_checkpoint('/path/to/ckpt')
parsed, = pydot.graph_from_dot_data(dot_string)
parsed.write_svg('/tmp/tensorflow/visualized_checkpoint.svg')

Example command line usage:

python -c "import tensorflow as tf;\
  print(tf.contrib.checkpoint.dot_graph_from_checkpoint('/path/to/ckpt'))"\
  | dot -Tsvg > /tmp/tensorflow/checkpoint_viz.svg

Args:

• save_path: The checkpoint prefix, as returned by tf.train.Checkpoint.save or tf.train.latest_checkpoint.

Returns:

A graph in DOT format as a string.