Add kernel SparseSegmentSum for CPU and WebGL backend

[ahmedsabie]

ref #4838
TensorFlow python version is defined here
c++ kernel definition (SparseSegmentReductionSumOp) is here

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

---

This change is 

[pyu10055]

Thanks, great work!

Reviewable status: 0 of 1 approvals obtained (waiting on @ahmedsabie and @lina128)

---

tfjs-backend-cpu/src/kernels/SparseSegmentReduction_impl.ts, line 22 at r1 (raw file):

export function sparseSegmentReductionImpl(
    input: TypedArray, inputShape: number[], inputDType: DataType,
    indices: TypedArray, segmentIds: TypedArray, isMean = false,

parameters isMean is not used ?

---

tfjs-backend-cpu/src/kernels/SparseSegmentReduction_impl.ts, line 61 at r1 (raw file):

  const outputLength =
      outputShape.reduce((product, value) => product * value, 1);
  const output = util.getArrayFromDType(inputDType, outputLength) as TypedArray;

Please comment that the output type array is initialized with 0 by default.

---

tfjs-backend-cpu/src/kernels/SparseSegmentReduction_impl.ts, line 117 at r1 (raw file):

      }
    }
    // const badOffset =

removed here?

---

tfjs-core/src/ops/sparse/sparse_segment_sum.ts, line 32 at r1 (raw file):

 * // Select two rows, one segment.
 * const result1 = tf.sparse.sparseSegmentSum(c,
 *                                           tf.tensor1d([0, 1], 'int32),

missing right ' across all examples

---

tfjs-core/src/ops/sparse/sparse_segment_sum.ts, line 34 at r1 (raw file):

 *                                           tf.tensor1d([0, 1], 'int32),
 *                                           tf.tensor1d([0, 0], 'int32);
 * console.log(result1);

console.log() is duplicated with tensor.print() method`

---

tfjs-core/src/ops/sparse/sparse_segment_sum.ts, line 35 at r1 (raw file):

 *                                           tf.tensor1d([0, 0], 'int32);
 * console.log(result1);
 * result1['output'].print(); // [[0 0 0 0]]

the method only returns a single tensor, it should be result1.print()

---

tfjs-core/src/ops/sparse/sparse_segment_sum.ts, line 67 at r1 (raw file):

      convertToTensor(segmentIds, 'segmentIds', 'sparseSegmentSum');

  if ($data.rank < 1) {

does data need to be 2D tensor or at least 1D tensor? The jsDoc says 2D?

---

tfjs-core/src/ops/sparse/sparse_segment_sum_test.ts, line 38 at r1 (raw file):

}

// function sparseTensorValue2x3x4() {

delete this method if it is not used.

---

tfjs-core/src/ops/sparse/sparse_segment_sum_test.ts, line 69 at r1 (raw file):

    const result =
        tf.sparse.sparseSegmentSum(TensorValue10(), [8, 3, 0, 9], [0, 1, 2, 2]);
    expectArraysClose(await result.data(), [9, 4, 11]);

should this be [9, 4, 10]?

---

tfjs-core/src/ops/sparse/sparse_segment_sum_test.ts, line 148 at r1 (raw file):

  });

  it('segments invalid 7', async () => {

can you add tests for invalid input ranks?

[ahmedsabie]

Reviewable status: 0 of 1 approvals obtained (waiting on @ahmedsabie, @lina128, and @pyu10055)

---

tfjs-backend-cpu/src/kernels/SparseSegmentReduction_impl.ts, line 22 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

parameters isMean is not used ?

good catch I will add it back in for the SparseSegmentMean

---

tfjs-core/src/ops/sparse/sparse_segment_sum.ts, line 67 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

does data need to be 2D tensor or at least 1D tensor? The jsDoc says 2D?

fixed it should be at least 1D

---

tfjs-core/src/ops/sparse/sparse_segment_sum_test.ts, line 69 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

should this be [9, 4, 10]?

the array has values 1 through 10 so indices 0 and 9 in segment 2 means value is 1 + 10 i believe

---

tfjs-core/src/ops/sparse/sparse_segment_sum_test.ts, line 148 at r1 (raw file):

Previously, pyu10055 (Ping Yu) wrote…

can you add tests for invalid input ranks?

done

[pyu10055]

Thanks!

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @ahmedsabie and @lina128)

---

tfjs-core/src/ops/sparse/sparse_segment_sum_test.ts, line 69 at r1 (raw file):

Previously, ahmedsabie wrote…

the array has values 1 through 10 so indices 0 and 9 in segment 2 means value is 1 + 10 i believe

got it

[pyu10055]

Looks like you need to disable the tests for WASM

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @ahmedsabie and @lina128)