env_specs: env_name: 'halfcheetah' env_kwargs: {} # kwargs to pass to the env constructor call

Should call this FIRST THING in the __init__ method if you ever want to serialize or clone this network. Usage: ```

Set the seed for all the possible random number generators. :param seed: :return: None

Close all of the environments. This function will be called only once (if not, it will be called during garbage collected). This way,

env_specs: env_name: 'halfcheetah' env_kwargs: {} # kwargs to pass to the env constructor call

Set up logger to have some reasonable default settings. Will save log output to based_log_dir/exp_prefix/exp_name. exp_name wi

Eval this module with a numpy interface Same as a call to __call__ except all Variable input/outputs are replaced with numpy

Return a batch of size `batch_size`. :param batch_size: :return:

Implement anything that needs to happen after every step under vec envs :return:

Implement anything that needs to happen after every vec env rollout.

Add a path to the replay buffer. This default implementation naively goes through every step, but you may want to optimize t

Remove invisible ANSI color codes.

Return a string which represents a horizontal line.

Return a string which represents a row of data cells.

Add a transition tuple.

Probably will be overridden by each algorithm

Normalize observations by statistics in obs_rms.

Let the replay buffer know that the episode has terminated in case some special book-keeping has to happen. :return:

deterministic=False makes no diff, just doing this for consistency in interface for now

:param observation: :return: action, debug_dictionary

:param value: some value, x :param pre_tanh_value: arctanh(x) :return:

Format a fixed width table for pretty printing. >>> print(tabulate([[1, 2.34], [-56, "8.999"], ["2", "10001"]])) --- --------- 1

Set training mode to `mode`. :param mode: If True, training will happen (e.g. set the dropout probabilities to not all ones).

Format row according to DataRow format without padding.

One annoying thing about the logger table is that the keys at each iteration need to be the exact same. So unless you can compute