utils.read_experiment

utils.read_experiment(
    log_path,
    parameter_set=None,
    job_descriptor='',
    iteration_number=None,
    summary_keys=('train_episode_returns', 'eval_episode_returns'),
    verbose=False
)

Reads in a set of experimental results from log_path.

The provided parameter_set is an ordered_dict which 1) defines the parameters of
this experiment, 2) defines the order in which they occur in the job descriptor.

The method reads all experiments of the form

${log_path}/${job_descriptor}.format(params)/logs,

where params is constructed from the cross product of the elements in the
parameter_set.

For example: parameterset = collections.OrderedDict([ (‘game’, [‘Asterix’,
‘Pong’]), (‘epsilon’, [‘0’, ‘0.1’]) ]) read_experiment(‘/tmp/logs’,
parameter_set, job_descriptor=’{}{}’) Will try to read logs from: -
/tmp/logs/Asterix_0/logs - /tmp/logs/Asterix_0.1/logs - /tmp/logs/Pong_0/logs -
/tmp/logs/Pong_0.1/logs

Args:

• log_path: string, base path specifying where results live.
• parameter_set: An ordered_dict mapping parameter names to allowable
  values.
• job_descriptor: A job descriptor string which is used to construct
  the full path for each trial within an experiment.
• iteration_number: Int, if not None determines the iteration number
  at which we read in results.
• summary_keys: Iterable of strings, iteration statistics to
  summarize.
• verbose: If True, print out additional information.

Returns:

A Pandas dataframe containing experimental results.