experiment Package

class explauto.experiment.experiment.Experiment(environment, agent)

Bases: explauto.utils.observer.Observer

This class is used to setup, run and log the results of an experiment.

Parameters:

• environment (Environment) – an environment
• agent (Agent) – an agent

run(n_iter=-1, bg=False)

Run the experiment.

Parameters:

• n_iter (int) – Number of run iterations, by default will run until the last evaluation step.
• bg (bool) – whether to run in background (using a Thread)

wait()

Wait for the end of the run of the experiment.

stop()

Stop the experiment.

fast_forward(log)

evaluate_at(eval_at, testcases)

Sets the evaluation interation indices.

Parameters:

• eval_at (list) – iteration indices where an evaluation should be performed
• testcases (numpy.array) – testcases used for evaluation

classmethod from_settings(settings)

Creates a Experiment object thanks to the given settings.

class explauto.experiment.pool.ExperimentPool(settings, evaluate_at, testcases=None, same_testcases=False)

Bases: object

Pool of experiments running in parallel.

Parameters:

• settings (list) – list of settings used to create the pool of experiments.
• evaluate_at (list) – iteration indices where an evaluation should be performed.
• testcases (numpy.array) – specify a pre-defined testcases, by default a testcases will be generated.
• same_testcases (bool) – whether to use the same testcases for all experiments.

The Pool will create Experiment using the from_settings() constructor for each combination of parameters given.

Note

If you set same_testcases to True the first experiment will generate a testcase used by all the others experiment. Otherwise, each experiment will generate its own testcase.

classmethod from_settings_product(environments, babblings, interest_models, sensorimotor_models, evaluate_at, testcases=None, same_testcases=False)

Creates a ExperimentPool with the product of all the given settings.

Parameters:

• environments (list of (environment name, config name)) – e.g. [(‘simple_arm’, ‘default’), (‘simple_arm’, ‘high_dimensional’)]
• babblings – e.g. [‘motor’, ‘goal’]
• interest_models (list of (interest model name, config name)) – e.g. [(‘random’, ‘default’)]
• sensorimotor_models (list of (sensorimotor model name, config name)) – e.g. [(‘non_parametric’, ‘default’)]
• evaluate_at (list of int) – indices defining when to evaluate
• same_testcases (bool) – whether to use the same testcases for all experiments

run(repeat=1, processes=None, use_thread=False)

Runs all experiments using a multiprocessing.Pool.

Parameters:

• repeat (int) – Number of time each experiment will be repeated.
• processes (int) – Number of processes launched in parallel (Default: uses all the availabled CPUs)
• use_thread (bool) – Use a ThreadPool instead of a Pool. By default, tries to guess depending on the environment which one to use.

settings

Returns a copy of the list of all the settings used.

logs

Returns the list of ExperimentLog.

Note

The logs will be returned as a vector if repeat was set as one in the run() method else it will be a matrix where each rows represents the n repetitions of an experiment.

class explauto.experiment.log.ExperimentLog(conf, expl_dims, inf_dims)

Bases: object

logs

add(topic, message)

pack(topic_dims, t)

Packs selected logs into a numpy array :param list topic_dims: list of (topic, dims) tuples, where topic is a string and dims a list dimensions to be packed for that topic :param int t: time indexes to be packed

axes_limits(topic_dims)

data_t(topic_dims, t)

scatter_plot(ax, topic_dims, t=None, ms_limits=True, **kwargs_plot)

2D or 3D scatter plot.

Parameters:

• ax (axes) – matplotlib axes (use Axes3D if 3D data)
• topic_dims (tuple) – list of (topic, dims) tuples, where topic is a string and dims is a list of dimensions to be plotted for that topic.
• t (int) – time indexes to be plotted
• kwargs_plot (dict) – argument to be passed to matplotlib’s plot function, e.g. the style of the plotted points ‘or’
• ms_limits (bool) – if set to True, automatically set axes boundaries to the sensorimotor boundaries (default: True)

plot_learning_curve(ax, squared_errors=False)

density_plot(ax, topic_dims, t=None, res_x=40, res_y=40, width_x=4, width_y=4)