Tensor stats migration

[darshil929]

Changes

Reason for changes

Related tickets

Tests

[darshil929]

Hi @kshpv @alexsu52

This PR is related to the issue #3041 and the changes implemented are:

• Removed circular dependencies between old and new implementations
• Migrated statistics classes imports to use experimental versions
• Updated tensor collector references to use the experimental implementation
• Updated PyTorch and OpenVINO implementations
• Added certain comments for my reference which will be removed during the final cleanup if necessary

Current status:

• Old implementation is now importing experimental TensorStatistic
• Framework-specific implementations are partially updated

I noticed the experimental implementation is missing:

• statistic_point.py (StatisticPoint and StatisticPointsContainer)
• aggregator.py (StatisticsAggregator)
• statistics_serializer.py and statistics_validator.py

These components are widely used throughout the codebase (30+ files).

Could you please advice if I should implement these missing components as part of the experimental implementation, or is there a different approach you would recommend?

[darshil929]

Hi @alexsu52

Sir, could you please have a look at the previous comment and advice me on how I should proceed?

[alexsu52]

Will this help with cyclical dependencies?

Suggested change

	AggregationAxes = Tuple[int, ...]
	AggregationAxes = Tuple[int, ...]
	ReductionAxes = Tuple[int, ...]

[alexsu52]

Hi @kshpv @alexsu52

This PR is related to the issue #3041 and the changes implemented are:

* Removed circular dependencies between old and new implementations

* Migrated statistics classes imports to use experimental versions

* Updated tensor collector references to use the experimental implementation

* Updated PyTorch and OpenVINO implementations

* Added certain comments for my reference which will be removed during the final cleanup if necessary

Current status:

* Old implementation is now importing experimental TensorStatistic

* Framework-specific implementations are partially updated

I noticed the experimental implementation is missing:

* `statistic_point.py` (StatisticPoint and StatisticPointsContainer)

* `aggregator.py` (StatisticsAggregator)

* `statistics_serializer.py` and `statistics_validator.py`

These components are widely used throughout the codebase (30+ files).

Could you please advice if I should implement these missing components as part of the experimental implementation, or is there a different approach you would recommend?

statistic_point.py (StatisticPoint and StatisticPointsContainer), aggregator.py (StatisticsAggregator), statistics_serializer.py and statistics_validator.py supports statistics from the experimental module. You should replace the following imports on statistics from the experimental module:

[daniil-lyakhov]

Hello, @darshil929

Thank you for your contribution! Unfortunately this ticket was recently closed by me here #3761

Anyway, thank you for your help and time!