Datasets¶

dataset_slice ¶

dataset_slice(dataset, start, end)

Return a shallow slice of a FeatureDataset.

Parameters:

Name	Type	Description	Default
`dataset`	`FeatureDataset`	Dataset to slice.	required
`start`	`int`	Starting index (inclusive).	required
`end`	`int`	Ending index (exclusive).	required

Returns:

Type	Description
`FeatureDataset`	Shallow slice of the dataset.

summarize_predictions_chunked(
    model_apply,
    model_params,
    features,
    param_names,
    *,
    chunk_size,
    param_shapes,
)

Compute prediction mean without materializing the full tensor.

Parameters:

Name	Type	Description	Default
`model_apply`	`Callable`	Model apply function.	required
`model_params`	`object`	Model parameters.	required
`features`	`ndarray`	Feature matrix used for inference.	required
`param_names`	`Sequence[str]`	Parameter names corresponding to predicted vector entries.	required
`chunk_size`	`int`	Batch size for chunked processing.	required
`param_shapes`	`Mapping[str, Sequence[int]]`	Target shapes keyed by parameter name.	required

Returns:

Type	Description
`Dict[str, object]`	Mean prediction mapped into structured Python scalars/arrays.

Raises:

Type	Description
`ValueError`	If `features` is empty.

truncate_dataset_timesteps(dataset, max_timesteps)

Trim controls/targets to the provided horizon (no-op if already shorter).

Parameters:

Name	Type	Description	Default
`dataset`	`FeatureDataset`	Dataset to truncate.	required
`max_timesteps`	`int`	Maximum number of timesteps to retain.	required

Returns:

Type	Description
`FeatureDataset`	Dataset with controls/targets truncated to `max_timesteps`.

Raises:

Type	Description
`ValueError`	If `max_timesteps` is not positive.