tf.compat.v1.profiler.Profiler

View source on GitHub

TensorFlow multi-step profiler.

https://github.com/tensorflow/tensorflow/tree/master/tensorflow/core/profiler/README.md

Typical use case:
  # Currently we are only allowed to create 1 profiler per process.
  profiler = Profiler(sess.graph)

  for i in xrange(total_steps):
    if i % 10000 == 0:
      run_meta = tf.compat.v1.RunMetadata()
      _ = sess.run(...,
                   options=tf.compat.v1.RunOptions(
                       trace_level=tf.RunOptions.FULL_TRACE),
                   run_metadata=run_meta)
      profiler.add_step(i, run_meta)

      # Profile the parameters of your model.
      profiler.profile_name_scope(options=(option_builder.ProfileOptionBuilder
          .trainable_variables_parameter()))

      # Or profile the timing of your model operations.
      opts = option_builder.ProfileOptionBuilder.time_and_memory()
      profiler.profile_operations(options=opts)

      # Or you can generate a timeline:
      opts = (option_builder.ProfileOptionBuilder(
              option_builder.ProfileOptionBuilder.time_and_memory())
              .with_step(i)
              .with_timeline_output(filename).build())
      profiler.profile_graph(options=opts)
    else:
      _ = sess.run(...)
  # Auto detect problems and generate advice.
  profiler.advise()

graph tf.Graph. If None and eager execution is not enabled, use default graph.
op_log optional. tensorflow::tfprof::OpLogProto proto. Used to define extra op types.

Methods

add_step

View source

Add statistics of a step.

Args
step int, An id used to group one or more different run_meta together. When profiling with the profile_xxx APIs, user can use the step id in the options to profile these run_meta together.
run_meta RunMetadata proto that contains statistics of a session run.

advise

View source

Automatically detect problems and generate reports.

Args
options A dict of options. See ALL_ADVICE example above.

Returns
An Advise proto that contains the reports from all checkers.

profile_graph

View source

Profile the statistics of graph nodes, organized by dataflow graph.

Args
options A dict of options. See core/profiler/g3doc/options.md.

Returns
a GraphNodeProto that records the results.

profile_name_scope

View source

Profile the statistics of graph nodes, organized by name scope.

Args
options A dict of options. See core/profiler/g3doc/options.md.

Returns
a GraphNodeProto that records the results.

profile_operations

View source

Profile the statistics of the Operation types (e.g. MatMul, Conv2D).

Args
options A dict of options. See core/profiler/g3doc/options.md.

Returns
a MultiGraphNodeProto that records the results.

profile_python

View source

Profile the statistics of the Python codes.

By default, it shows the call stack from root. To avoid redundant output, you may use options to filter as below options['show_name_regexes'] = ['.my_code.py.']

Args
options A dict of options. See core/profiler/g3doc/options.md.

Returns
a MultiGraphNodeProto that records the results.

serialize_to_string

View source

Serialize the ProfileProto to a binary string.

Users can write it to file for offline analysis by tfprof commandline or graphical interface.

Returns
ProfileProto binary string.