Gradients will and should *not* pass through this operation. See https://github.com/pytorch/pytorch/issues/4620 for discussion.

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

:param latents: :return: reconstruction, obs_distribution_params

:param input: :return: latent_distribution_params

:param observation: :return: action, debug_dictionary

Remove invisible ANSI color codes.

Usage: ``` def foo(variant): x = variant['x'] y = variant['y'] logger.log("sum", x+y) variant = { 'x'

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

Return a string which represents a horizontal line.

Return a string which represents a row of data cells.

Use this function to register an array to be shared. This will wipe arr.

Iterate over the hyperparameters in a grid-manner. :return: List of dictionaries. Each dictionary is a map from name to hype

Plot two kinds of visualizations here: (1) Trend of loss_minus with respect to loss_plus (2) Histogram of different gradient di

Convolutional network used to compute the agent's Q-value distribution. Args: num_actions: int, number of actions. num_atoms: int, th

Update policy and then look at how the returns under this policy look like.

Turns list of dicts into dict of np arrays

:param latent_distribution_params: :return: kl div between latent_distribution_params and prior on latent space

Apply internal replay buffer representation changes: save images as bytes

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

:param latent_distribution_params: :return: latents

Run an experiment locally without any serialization. :param experiment_function: Function. `variant` will be passed in as its only argum

Data saved here will always override the last entry :param data: Something pickle'able.

Format row according to DataRow format without padding.

>>> _isint("123") True >>> _isint("123.45") False

>>> _isnumber("123.45") True >>> _isnumber("123") True >>> _isnumber("spam") False

This method will save all the replay buffer's state in a single file.

This method will save all the replay buffer's state in a single file.

Utility function for combining multiple Q functions. Args: q_functions: Multiple Q-functions concatenated. transform_strategy: str, Possibl