com.linkedin.feathr.common.tensor

Class TensorTypes

java.lang.Object

com.linkedin.feathr.common.tensor.TensorTypes

public final class TensorTypes
extends java.lang.Object

Utility functions for working with TensorType

Method Summary

Modifier and Type	Method and Description
static TensorType	fromRepresentables(boolean sparse, Representable[] representables) Creates a synthetic TensorType from the information available on TensorData.
static TensorType	fromTensorData(TensorData tensorData) Creates a synthetic TensorType from a TensorData.
static TensorType	parseTensorType(java.lang.String syntax) Creates a TensorType from a string representation of the type.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Detail

parseTensorType

public static TensorType parseTensorType(java.lang.String syntax)

Creates a TensorType from a string representation of the type. Intended to simplify creation of automated tests. It is NOT robust (e.g., the only whitespace allowed is the single mandatory space after comma). Some examples: TENSOR[STRING][LONG]:DOUBLE TENSOR[INT(3)]:FLOAT TENSOR:BOOLEAN

fromRepresentables

public static TensorType fromRepresentables(boolean sparse,
                                            Representable[] representables)

Creates a synthetic TensorType from the information available on TensorData. This is a stop-gap solution for integrations where consumer requires TensorType but the producer does not have it. Please use sparingly. NOTE that the static shape is not recoverable, so all dimensions in the TensorType will have the unknown shape. RAGGED tensors are not supported.

Parameters:

sparse - if the type should be sparse

representables - primitive types of the tensor's columns

Returns:

a TensorType corresponding to the inputs

fromTensorData

public static TensorType fromTensorData(TensorData tensorData)

Creates a synthetic TensorType from a TensorData. This is a stop-gap solution for integrations where consumer requires TensorType but the producer does not have it. Please use sparingly. NOTE that the static shape is not recoverable (for dense tensors the static shape can have some of the dimensions unknown, so it is not recoverable from the dynamic shape; sparse tensors do not have even the dynamic shape), so all dimensions in the TensorType will have the unknown shape. RAGGED tensors are not supported.

Parameters:

tensorData - an instance of tensor from which to derive type information

Returns:

a TensorType corresponding to the input