Request an attribute from the environment. Note that this involves communication with the external process, so it can be slow. Args:

Yields agent_id and agent for all agents in environment.

Ensure that no other worker will pick up this run or raise an exception. Note that usually the last worker who claims a run wins, since its name

Helper method for compiling the loss function. The loss function is obtained from the log likelihood, assuming that the output distri

Helper method for compiling the loss function. The loss function is obtained from the log likelihood, assuming that the output distri

Compiles the output of the network at the given inputs. If inputs is 2D, returns a 3D tensor where output[i] is the output of the ith network

Compiles the output of the network at the given inputs. If inputs is 2D, returns a 3D tensor where output[i] is the output of the ith network

Repeat training _n_train_repeat times every _train_every_n_steps

Construct TensorFlow feed_dict from sample batch.

Construct TensorFlow feed_dict from sample batch.

concat all agent's observation to construct state info

concat all agent's observation to construct state info

Initialize or restore variables from a checkpoint if available. Args: sess: Session to initialize variables in. savers: List of saver

Flatten consecutive axes of a tensor trying to preserve static shapes.

Get a tensor shape from a OpenAI Gym space. Args: space: Gym space. Raises: NotImplementedError: For spaces other than Box and D

Mark run as finished by writing a DONE file.

Yields position, number of agent and tree strength for all cells defined by `mask`. Args: mask: tuple of slices in extended coord

Convert an action number to a string representing the action. Args: action_number: an integer from the action space.

Add a phase to the trainer protocol. The score tensor can either be a scalar or vector, to support single and batched computations. Args