Add specification for SegmentMax-16

[p-wysocki]

Details:

• Specification for tf.math.segment_max (https://www.tensorflow.org/api_docs/python/tf/math/segment_max)

Tickets:

• CVS-158914

[rkazants]

Let us have EmbeddingSegmentsMax similar to EmbeddingSegmentsSum.
It should also have default index (defining default value for empty segment)

[rkazants]

we should have default value for empty segments, otherwise, we will have additional computation graph (that is not trivial) to compute empty segments and replace zero value

[p-wysocki]

The default value seems to be 0, according to https://www.tensorflow.org/api_docs/python/tf/raw_ops/SegmentMax. I don't think we should expand the op on our own, especially since we only expect it to come from TF FE.

[mitruska]

There is also V2 https://www.tensorflow.org/api_docs/python/tf/raw_ops/SegmentMaxV2 where the default has been changed to numeric_limits<T>::lowest(). Adding attribute for default value seems to be a simple solution to support both cases, but to enable V2 at once we would also need to consider "num_segments" input.

[rkazants]

please define input shapes and output shape for each input and describe what dimensions are equal

[p-wysocki]

data can have any rank and dimensions, so it's described as ND of any numerical type. segment_ids are specified to be a 1D tensor of non-negative, sorted integer numbers of size equal to the size of the first dimension of the input tensor.

Could you please specify what's missing? I think the shapes are covered, but I may be missing something.

[p-wysocki]

Let us have EmbeddingSegmentsMax similar to EmbeddingSegmentsSum.

EmbeddingSegmentX is for sparse inputs, while SegmentMax I'm implementing (to unlock some models) accepts dense inputs, so I don't think it should be added as EmbeddingSegmentMax. I'm moving the discussion to internal channels, if it results in changes, I'll apply them to the PR.

[mitruska]

Suggested change

	* **1**: The output tensor of type *T* and the same shape as the ``input`` tensor with the exception for the first dimension, which is equal to the count of unique segment IDs.
	* **1**: The output tensor of type *T* and almost the same shape as the ``data`` input tensor with the exception for the first dimension, which is equal to the count of unique segment IDs (calculated as ``max(segment_ids) + 1``).

[praasz]

Maybe instead almost use,

The output tensor has same rank and dimensions as the ``data`` input tensor except first dimension which is calculated as ``max(segment_ids) + 1`` 

?

[mitruska]

There is also V2 https://www.tensorflow.org/api_docs/python/tf/raw_ops/SegmentMaxV2 where the default has been changed to numeric_limits<T>::lowest(). Adding attribute for default value seems to be a simple solution to support both cases, but to enable V2 at once we would also need to consider "num_segments" input.

[mitruska]

The style of the "Inputs" section description follows rather the "Attributes" style.
Consider alignment with other spec documents.

[mitruska]

I can see that unsorted segment_ids may lead to error or undefined behavior (implementation specific, depends on the hardware).
Should we specify a common behavior for OV op?
Can be clarified at the plugin implementation stage.

[mitruska]

Suggested change

	* **Range of values**: 1D tensor of non-negative, sorted integer numbers. Its size is equal to the size of the first dimension of the input tensor.
	* **Range of values**: 1D tensor of non-negative, sorted integer numbers. Its size is equal to the size of the first dimension of the ``data`` input tensor.

[praasz]

Suggested change

	SegmentMax
	===================
	SegmentMax
	==========

Should number = be same as heading length?

[praasz]

Maybe instead almost use,

The output tensor has same rank and dimensions as the ``data`` input tensor except first dimension which is calculated as ``max(segment_ids) + 1`` 

?