lale.lib.rasl.batching module

class lale.lib.rasl.batching.Batching(*, operator, batch_size=64, shuffle=False, num_workers=0, inmemory=False, num_epochs=None, max_resident=None, scoring=None, progress_callback=None, partial_transform=False, priority='resource_aware', verbose=0)

Bases: PlannedIndividualOp

Batching trains the given pipeline using batches.

This documentation is auto-generated from JSON schemas.

The batch_size is used across all steps of the pipeline, serializing the intermediate outputs if specified.

Parameters

• operator (operator, optional, not for optimizer) – A lale pipeline object to be used inside of batching

• batch_size (integer, >=1, >=32 for optimizer, <=128 for optimizer, uniform distribution, optional, default 64) – Batch size used for transform.

• shuffle (boolean, optional, not for optimizer, default False) – Shuffle dataset before batching or not.

• num_workers (integer, optional, not for optimizer, default 0) – Number of workers for pytorch dataloader.

• inmemory (boolean, optional, not for optimizer, default False) –

  Whether all the computations are done in memory

  or intermediate outputs are serialized. Only applies to transform/predict. For fit, use the max_resident argument.

• num_epochs (union type, optional, not for optimizer, default None) –

  Number of epochs. If the operator has num_epochs as a parameter, that takes precedence.

  • integer

  • or None

• max_resident (union type, optional, not for optimizer, default None) –

  Amount of memory to be used in bytes.

  • integer

  • or None

• scoring (union type, optional, not for optimizer, default None) –

  Batch-wise scoring metrics from lale.lib.rasl.

  • callable

  • or None

• progress_callback (union type, optional, not for optimizer, default None) –

  Callback function to get performance metrics per batch.

  • callable

  • or None

• partial_transform (boolean, optional, not for optimizer, default False) – Whether to allow partially-trained upstream operators to transform data for training downstream operators even before the upstream operator has been fully trained.

• priority (‘batch’, ‘step’, or ‘resource_aware’, optional, not for optimizer, default ‘resource_aware’) – Scheduling priority in task graphs. “batch” will execute tasks from earlier batches first. “step” will execute tasks from earlier steps first, like nested-loop algorithm. And “resource_aware” will execute tasks with less non-resident data first.

• verbose (integer, optional, not for optimizer, default 0) – Verbosity level, higher values mean more information.

fit(X, y=None, **fit_params)

Train the operator.

Note: The fit method is not available until this operator is trainable.

Once this method is available, it will have the following signature:

Parameters

• X (union type) –

  Features; the outer array is over samples.

  • array

    • items : union type

      • float

      • or string

      • or boolean

  • or array

    • items : array

      • items : union type

        • float

        • or string

        • or boolean

  • or dict

• y (union type) –

  • array

    • items : union type

      • integer

      • or float

      • or string

  • or None

• classes (union type, optional) –

  The total number of classes in the entire training dataset.

  • array

    • items : union type

      • float

      • or string

      • or boolean

  • or None

predict(X, **predict_params)

Make predictions.

Note: The predict method is not available until this operator is trained.

Once this method is available, it will have the following signature:

Parameters

• X (union type) –

  Features; the outer array is over samples.

  • array

    • items : union type

      • float

      • or string

      • or boolean

  • or array

    • items : array

      • items : union type

        • float

        • or string

        • or boolean

  • or any type

• y (array, optional) –

  • items : union type

    • integer

    • or float

Returns

result – Output data schema for transformed data.

Return type

Any

transform(X, y=None)

Transform the data.

Note: The transform method is not available until this operator is trained.

Once this method is available, it will have the following signature:

Parameters

• X (union type) –

  Features; the outer array is over samples.

  • array

    • items : union type

      • float

      • or string

      • or boolean

  • or array

    • items : array

      • items : union type

        • float

        • or string

        • or boolean

  • or any type

• y (array, optional) –

  • items : union type

    • integer

    • or float

Returns

result – Output data schema for transformed data.

Return type

Any