Computes gradients for SparseSegmentSum.
Returns tensor "output" with same shape as grad, except for dimension 0 whose value is the number of unique indexes in "indices". Also returns vector "sorted_unique_indices" containing the corresponding indexes from "indices".
Public Methods
static <T extends Number, U extends Number, V extends Number> SparseSegmentSumGradV2<T, U> | |
Output<T> |
output()
|
Output<U> |
Inherited Methods
Public Methods
public static SparseSegmentSumGradV2<T, U> create (Scope scope, Operand<T> grad, Operand<U> indices, Operand<V> segmentIds, Operand<Integer> denseOutputDim0)
Factory method to create a class wrapping a new SparseSegmentSumGradV2 operation.
Parameters
scope | current scope |
---|---|
grad | gradient propagated to the SparseSegmentSum op. |
indices | indices passed to the corresponding SparseSegmentSum op. |
segmentIds | segment_ids passed to the corresponding SparseSegmentSum op. |
denseOutputDim0 | dimension 0 of "data" passed to SparseSegmentSum op. |
Returns
- a new instance of SparseSegmentSumGradV2