Batching¶

batchify ¶

batchify(x, agent_list, num_actors)

Stack agent data into a batched array.

Parameters:

Name	Type	Description	Default
`x`	`dict`	Mapping from agent id to arrays.	required
`agent_list`	`Iterable`	Ordered agent identifiers.	required
`num_actors`	`int`	Number of actors to include.	required

Returns:

Type	Description
`ndarray`	Stacked array of shape `(num_actors, -1)`.

create_batched_params(
    base_params, batch_size, **param_overrides
)

Create batched parameters from a base Param object.

Parameters:

Name	Type	Description	Default
`base_params`	`Param`	Base Param object with scalar values.	required
`batch_size`	`int`	Number of samples in the batch.	required
`**param_overrides`		Parameter overrides as arrays of shape `[batch_size]`.	`{}`

Returns:

Type	Description
`Param`	Batched parameter object.

extract_param_at_index(batched_params, index)

Extract parameters for a single sample from batched parameters.

Parameters:

Name	Type	Description	Default
`batched_params`	`Param`	Batched parameter object.	required
`index`	`int`	Index to extract.	required

Returns:

Type	Description
`Param`	Scalar parameter object for the given index.

is_batched_params(params)

Check if a Param object contains batched parameters.

Parameters:

Name	Type	Description	Default
`params`	`Param`	Parameter object to test.	required

Returns:

Type	Description
`bool`	True if any field is an ndarray, otherwise False.

unbatchify(x, agent_list, num_envs, num_actors)

Split a batched array back into a dict of agents.

Parameters:

Name	Type	Description	Default
`x`	`ndarray`	Batched array of shape `(num_actors * num_envs, ...)`.	required
`agent_list`	`Iterable`	Ordered agent identifiers.	required
`num_envs`	`int`	Number of environments.	required
`num_actors`	`int`	Number of actors.	required

Returns:

Type	Description
`dict`	Mapping from agent id to unbatched arrays.