unilab.tasks.motion_tracking.common.tensor_runtime.EpisodeMetrics

class unilab.tasks.motion_tracking.common.tensor_runtime.EpisodeMetrics[source]

Bases: object

Device-resident current episode accumulators.

snapshot and finished_values return tensors; callers decide when a logging or IPC boundary is allowed to copy them to host.

Parameters:

Methods

__init__(rewards, lengths)

create(num_envs, device)

finished_values(done_mask)

reset(done_mask)

snapshot()

update(reward, done)

Attributes

rewards

lengths

rewards: Tensor
lengths: Tensor
classmethod create(num_envs, device)[source]
Parameters:
Return type:

EpisodeMetrics

update(reward, done)[source]
Parameters:
Return type:

None

finished_values(done_mask)[source]
Parameters:

done_mask (Tensor)

Return type:

Tensor

reset(done_mask)[source]
Parameters:

done_mask (Tensor)

Return type:

None

snapshot()[source]
Return type:

dict[str, Tensor]

__init__(rewards, lengths)
Parameters: