SpacesRecord

class maze.core.trajectory_recording.records.spaces_record.SpacesRecord(actor_id: ActorID, observation: Dict[str, numpy.ndarray | torch.Tensor] | None = None, action: Dict[str, numpy.ndarray | torch.Tensor] | None = None, reward: float | numpy.ndarray | torch.Tensor | None = None, done: bool | numpy.ndarray | torch.Tensor | None = None, info: Dict | None = None, next_observation: Dict[str, numpy.ndarray | torch.Tensor] | None = None, logits: Dict[str, numpy.ndarray] | None = None, discounted_return: float | numpy.ndarray | None = None, batch_shape: List[int] | None = None, policy_record: object | None = None, env_time: int | None = None)

Record of spaces (i.e., raw action, observation, and associated data) from a single sub-step.

action: Dict[str, numpy.ndarray | torch.Tensor] | None = None

Action recorded during the step.

actor_id: ActorID

ID of the actor for this step.

property agent_id: int

Sub-step key (i.e., the second part of the Actor ID) for this step.

batch_shape: List[int] | None = None

If the record is batched, this is the shape of the batch.

discounted_return: float | numpy.ndarray | None = None

Discounted return for this step.

done: bool | numpy.ndarray | torch.Tensor | None = None

Done flag recorded during the step.

env_time: int | None = None

The env time (t) of the env when recording the observation such that: (s_t, a_t, v_t) -> env step -> (r_t, done_t, info_t) is recorded.

info: Dict | None = None

Info dictionary recorded during the step.

logits: Dict[str, numpy.ndarray] | None = None

Action logits recorded during the step.

next_observation: Dict[str, numpy.ndarray | torch.Tensor] | None = None

Observation obtained after this step (i.e., results of the action taken in this step).

observation: Dict[str, numpy.ndarray | torch.Tensor] | None = None

Observation recorded during the step.

policy_record: object | None = None

Policy specific data that can be recorded with the help of the write_policy_record method of the policy.

reward: float | numpy.ndarray | torch.Tensor | None = None

Reward recorded during the step.

classmethod stack(records: List[SpacesRecord]) → SpacesRecord

Stack multiple records into a single spaces record. Useful for processing multiple records in a batch.

All the records should be in numpy and have the same structure of the spaces (i.e. come from the same environment etc.).

Parameters:

records – Records to stack.

Returns:

Single stacked record, containing all the given records, and having the corresponding batch shape.

property substep_key: str | int

Sub-step key (i.e., the first part of the Actor ID) for this step.

to_numpy() → SpacesRecord

Convert the record to numpy.

to_torch(device: str) → SpacesRecord

Convert the record to Torch.

Parameters:

device – Device to move the tensors to.