Experiment

Experiment

A class to run an experiment with a given agent and environment.

Parameters:

Name	Type	Description	Default
name	str	The name of the experiment.	required
agent	Agent	The agent to use in the experiment.	required
env	Environment	The environment to use in the experiment.	required
env_id	str	The ID of the environment.	''
seeds	Tuple[int, ...]	The seeds to use in the experiment.	(0)
group	str	The group to use in the experiment.	''

Attributes:

Name	Type	Description
name	str	The name of the experiment.
agent	Agent	The agent to use in the experiment.
env	Environment	The environment to use in the experiment.
env_id	str	The ID of the environment.
seeds	Tuple[int, ...]	The seeds to use in the experiment.
group	str	The group to use in the experiment.

run(do_log=True)

Default function to run the experiment. This function compiles the training function, trains the agent, and logs the results.

Parameters:

Name	Type	Description	Default
do_log	bool	Whether to log the results to wandb.	True

Warning

Logging to wandb is usually much slower than training the agent itself. The time is linear in the number of seeds.

Returns:

Name	Type	Description
Tuple		A tuple containing the final training state and the logs.

run_hparam_search(hparams_distr, pop_size)

Function to run a hyperparameter search for the experiment. This function samples hyperparameters from the given distributions, trains the agent, and logs the results.

Parameters:

Name	Type	Description	Default
hparams_distr	Dict[str, Distribution]	A dictionary of hyperparameter distributions. The keys are the hyperparameter names, which must exist in self.agent.hparams, and the values are the corresponding distributions.	required
pop_size	int	The number of hyperparameter sets to sample.	required

Returns:

Name	Type	Description
Tuple		A tuple containing the final training states and the logs, batched over the hyperparameter sets.