mypy checks for common graph

[DaniAffCH]

Changes

This PR closes issue #2495 by addressing various mypy checks and enhancing type safety in the codebase.

• Resolved specific mypy errors related to type inconsistencies.
• Utilized # type:ignore for cases requiring significant refactoring due to untyped packages like networkx.
• Added the directory nncf/common/graph to .mypy.ini to include additional files for type checking.

Related Tickets

N/A

Tests

Pytests were run to ensure that the changes did not modify the common logic and that the codebase remained functional.

[DaniAffCH]

Actually, 3 mypy errors are remaining. I wanted to discuss them with you.

[DaniAffCH]

https://github.com/openvinotoolkit/nncf/blob/b7ba5ad1f8c785e11d3b2a3864bd33aa1787bb56/nncf/common/graph/patterns/manager.py#L118C1-L131C97

In this function, the parameter model_type is optional. However, the function _get_full_pattern_graph expects model_type to be of type ModelType and not optional. The case where model_type is None is not explicitly handled. Is this a problem? (I didn't modify the datatype because I wanted to discuss with you this detail before, to figure out if I should modify get_full_ignored_pattern_graph or _get_full_pattern_graph)

[DaniAffCH]

https://github.com/openvinotoolkit/nncf/blob/b7ba5ad1f8c785e11d3b2a3864bd33aa1787bb56/nncf/common/graph/graph.py#L677C1-L681C30

Here nx_edge[2] is accessed. However, the variable boundary comes from the function _get_edge_boundaries that returns a list of Tuple[str, str] with only 2 elements. If so, nx_edge[2] would be out of bounds.
Is the return type of _get_edge_boundaries incorrect, or is there a more fundamental issue at play?

[codecov]

Codecov Report

Attention: Patch coverage is 94.31818% with 5 lines in your changes are missing coverage. Please review.

Project coverage is 84.91%. Comparing base (b7ba5ad) to head (069a607).
Report is 4 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #2600      +/-   ##
===========================================
- Coverage    91.16%   84.91%   -6.26%     
===========================================
  Files          494      494              
  Lines        45350    45390      +40     
===========================================
- Hits         41342    38541    -2801     
- Misses        4008     6849    +2841     

Files	Coverage Δ
nncf/common/graph/graph_matching.py	100.00% <100.00%> (ø)
nncf/common/graph/layer_attributes.py	96.29% <100.00%> (ø)
nncf/common/graph/model_transformer.py	87.50% <100.00%> (ø)
nncf/common/graph/operator_metatypes.py	93.97% <100.00%> (ø)
nncf/common/graph/patterns/patterns.py	96.46% <100.00%> (+0.03%)	⬆️
nncf/common/graph/transformations/commands.py	97.01% <100.00%> (ø)
nncf/common/graph/transformations/layout.py	100.00% <100.00%> (ø)
nncf/common/graph/utils.py	82.14% <100.00%> (ø)
nncf/common/utils/registry.py	90.90% <100.00%> (ø)
nncf/common/graph/graph.py	96.25% <97.50%> (-0.27%)	⬇️
... and 1 more

... and 60 files with indirect coverage changes

Flag	Coverage Δ
COMMON	44.13% <69.31%> (-0.02%)	⬇️
ONNX	34.67% <78.40%> (+0.01%)	⬆️
OPENVINO	∅ <ø> (∅)
TENSORFLOW	30.10% <68.18%> (-0.02%)	⬇️
TORCH	65.92% <86.36%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
common	93.13% <94.31%> (-0.67%)	⬇️
torch	93.48% <ø> (+<0.01%)	⬆️
tensorflow	93.74% <ø> (ø)
onnx	93.09% <ø> (+0.06%)	⬆️
openvino	25.75% <ø> (-68.39%)	⬇️
ptq	69.92% <ø> (-20.20%)	⬇️

[alexsu52]

get_full_ignored_pattern_graph or _get_full_pattern_graph)

Thanks for your find. This is exactly what we expect from a mypy task. model_type can has None value

nncf/nncf/quantization/quantize_model.py

Line 101 in c79111b

model_type: Optional[ModelType] = None,

Thus typing of _get_full_pattern_graph and _filter_patterns should be modified.

[alexsu52]

https://github.com/openvinotoolkit/nncf/blob/b7ba5ad1f8c785e11d3b2a3864bd33aa1787bb56/nncf/common/graph/graph.py#L677C1-L681C30

Here nx_edge[2] is accessed. However, the variable boundary comes from the function _get_edge_boundaries that returns a list of Tuple[str, str] with only 2 elements. If so, nx_edge[2] would be out of bounds. Is the return type of _get_edge_boundaries incorrect, or is there a more fundamental issue at play?

It looks like the return typing of _get_edge_boundaries is wrong, nx.edge_boundary(graph, match, data=True) returns tuple from 3 elements:

nncf/nncf/common/graph/graph.py

Line 297 in b7ba5ad

out_edge_boundary = list(nx.edge_boundary(graph, match, data=True))

More information is here https://networkx.org/documentation/stable/reference/classes/generated/networkx.MultiGraph.edges.html

[DaniAffCH]

Thank you @alexsu52 for the clarification. All the mypy checks pass now, so the PR is ready to be reviewed.

[andrey-churkin]

Thanks for the contribution!