Recursively convert test/rew from all tfevent file under root_dir to csv. This function assumes that there is at most one tfevents file in each d

Initialize the logger for each episode.

Log information for each episode.

Evaluate the model

Initialize the logger for evaluation.

Log evaluation information.

Log evaluation information per step.

Log evaluation information.

Training data generator for critic that uses MLP network. Args: critic_num_mini_batch: (int) Number of mini batches for critic.

Flatten the hessians.

Construct the action for agent a_id.

Returns the available actions for agent_id

Returns the available actions for agent_id

Get the combined dimension of central observation and individual actions.

Load config file for user-specified algo and env. Args: algo: (str) Algorithm name. env: (str) Environment name. Returns:

Return RGB images from each environment

Get available joint idx for an agent. All other agents keep their current actions, and this agent can freely choose. Args:

Identify all joints at distance <= k from agent agent_id :param agent_id: id of agent to be considered :param agent_partitions: list of joint

Convert simulation states into the format of observation_space ------ Returns: (np.ndarray) - ego info

Returns the shape of the observation

construct baseline observation from task observation Baseline observation: # 0. ego delta altitude (unit: 1km)

construct baseline observation from task observation Baseline observation: # 0. ego delta altitude (unit: 1km)

Compute the reward at the current timestep. Overwritten by subclasses. Args: task: task instance env: environ

Returns the shape of the state

Get the task name.

Returns the total number of actions an agent could ever take

Returns the total number of actions an agent could ever take

Draw a sample from the Gumbel-Softmax distribution

Given an angle in degrees, normalises in (-180, 180]

Initialize the logger.

Initialise the units.

Insert data into buffer.

Check if v is JSON serializable.

KL divergence between two distributions.

Load action space

Load observation space

Main function.

Merge result in csv_files into a single csv file.

Training data generator for critic that uses RNN network. This generator does not split the trajectories into chunks, and therefore ma

Get next indices

Get next indices

Convert discrete action index into continuous value.

Convert discrete action index into continuous value.