SegmentMaxV2

Computes the maximum along segments of a tensor.

Read [the section on segmentation](https://tensorflow.org/api_docs/python/tf/math#Segmentation) for an explanation of segments.

Computes a tensor such that \\(output_i = \max_j(data_j)\\) where `max` is over `j` such that `segment_ids[j] == i`.

If the maximum is empty for a given segment ID `i`, it outputs the smallest possible value for the specific numeric type, `output[i] = numeric_limits::lowest()`.

Note: That this op is currently only supported with jit_compile=True.

Caution: On CPU, values in `segment_ids` are always validated to be sorted, and an error is thrown for indices that are not increasing. On GPU, this does not throw an error for unsorted indices. On GPU, out-of-order indices result in safe but unspecified behavior, which may include treating out-of-order indices as the same as a smaller following index.

The only difference with SegmentMax is the additional input `num_segments`. This helps in evaluating the output shape in compile time. `num_segments` should be consistent with segment_ids. e.g. Max(segment_ids) should be equal to `num_segments` - 1 for a 1-d segment_ids With inconsistent num_segments, the op still runs. only difference is, the output takes the size of num_segments irrespective of size of segment_ids and data. for num_segments less than expected output size, the last elements are ignored for num_segments more than the expected output size, last elements are assigned smallest possible value for the specific numeric type.

For example:

>>> @tf.function(jit_compile=True) ... def test(c): ... return tf.raw_ops.SegmentMaxV2(data=c, segment_ids=tf.constant([0, 0, 1]), num_segments=2) >>> c = tf.constant([[1,2,3,4], [4, 3, 2, 1], [5,6,7,8]]) >>> test(c).numpy() array([[4, 3, 3, 4], [5, 6, 7, 8]], dtype=int32)