Temporal Semantics#

TemporalAxis#

Use TemporalAxis inside element_names to mark one tensor axis as temporal and record the period between samples on that axis. This is useful for chunked outputs such as an action tensor shaped [num_chunks, num_joints].

Place TemporalAxis directly in the outer element_names list. Do not wrap it in a list. LEAPP serializes the temporal axis as the reserved __temporal_axis__ sentinel and emits temporal_period_ms as a sibling field on the tensor entry.

from leapp import TemporalAxis

TensorSemantics(
    "actions",
    actions,
    kind=OutputKindEnum.JOINT_TORQUES,
    element_names=[
        TemporalAxis(period_ms=100),
        ["hip", "knee", "ankle"],
    ],
)

- name: actions
  dtype: float32
  shape: [4, 3]
  type: tensor
  kind: target/joint/torques
  element_names:
  - __temporal_axis__
  - [hip, knee, ankle]
  temporal_period_ms: 100

Downstream consumers can find the temporal axis by locating __temporal_axis__ in element_names. LEAPP does not validate temporal_period_ms against GraphConfigs.frequency.

Warning

__temporal_axis__ is reserved for LEAPP output. Use TemporalAxis in Python annotations rather than writing the sentinel directly. A tensor may contain at most one temporal axis marker.

See Kind & Element Names for kind and element_names metadata.

On this page