[ruff] Enable pydoc rules (#3288)

### Changes

Enable rule https://docs.astral.sh/ruff/rules/#pydocstyle-d

ignored rules:
```
  "D100", # undocumented-public-module
  "D101", # undocumented-public-class
  "D102", # undocumented-public-method
  "D103", # undocumented-public-function
  "D104", # undocumented-public-package
  "D105", # undocumented-magic-method
  "D106", # undocumented-public-nested-class
  "D107", # undocumented-public-init
  "D200", # unnecessary-multiline-docstring
  "D203", # incorrect-blank-line-before-class
  "D205", # missing-blank-line-after-summary
  "D212", # multi-line-summary-first-line
  "D400", # missing-trailing-period
  "D401", # non-imperative-mood
  "D402", # signature-in-docstring
  "D404", # docstring-starts-with-this
  "D413", # missing-blank-line-after-last-section
  "D415", # missing-terminal-punctuation
  "D417", # undocumented-param
```
Some rules is not supported because used not supported style or require
add or change documentation in too many files

Author: AlexanderDokuchaev

nncf/common/composite_compression.py

@@ -75,7 +75,6 @@ def calculate(self, *args: Any, **kwargs: Any) -> Any:
 
         :return: The compression loss value.
         """
-
         if len(self._child_losses) == 0:
             msg = "Cannot calculate the loss value because the number of child loss is 0."
             raise nncf.InternalError(msg)

nncf/common/graph/graph.py

@@ -343,7 +343,6 @@ def get_previous_nodes(self, node: NNCFNode) -> List[NNCFNode]:
         :param node: Consumer node.
         :return: List of producers nodes of provided node.
         """
-
         nx_node_keys = self._nx_graph.pred[self._node_id_to_key_dict[node.node_id]]
         return [self._nodes[key] for key in nx_node_keys]
 
@@ -723,7 +722,6 @@ def get_nncf_graph_pattern_io(self, match: List[str]) -> NNCFGraphPatternIO:
         `match` list
         :return: NNCFGraphPatternIO object describing the inputs and outputs of the matched subgraph
         """
-
         in_edge_boundary, out_edge_boundary = NNCFGraph._get_edge_boundaries(match, self._nx_graph)
         boundary = in_edge_boundary + out_edge_boundary
         input_nncf_edges = []

nncf/common/graph/graph_matching.py

@@ -57,6 +57,7 @@ def _is_subgraph_matching_strict(graph: nx.DiGraph, pattern: nx.DiGraph, subgrap
     3) External successors or predecessors of the nodes which are not starting and last.
     If any of these conditions is True, than returns False, otherwise - True.
     The checks are skipped for NON_PATTERN_NODE_TYPE.
+
     Example:
     This subgraph matching is not strict.
     (conv2d + BN + ReLU pattern):
@@ -71,6 +72,7 @@ def _is_subgraph_matching_strict(graph: nx.DiGraph, pattern: nx.DiGraph, subgrap
            (cat)----/
              |
             ...
+
     :param graph: The model graph.
     :param pattern: The matched pattern.
     :param subgraph: A subgraph of the model graph including the nodes outside the pattern.

nncf/common/graph/patterns/patterns.py

@@ -101,7 +101,6 @@ def __add__(self, other: "GraphPattern") -> "GraphPattern":
         :param other: GraphPattern that will be added.
         :return: resulted GraphPattern.
         """
-
         final_pattern = GraphPattern()
         for self_subgraph in self.get_weakly_connected_subgraphs():
             for other_subgraph in other.get_weakly_connected_subgraphs():

nncf/common/insertion_point_graph.py

@@ -87,7 +87,6 @@ def __init__(
         If left unspecified, every node in `nncf_graph` will be allowed to have a single post-hook for its output
          (post-hooking separate tensors in an operation's output is not currently supported)
         """
-
         super().__init__()
         self._base_nx_graph = deepcopy(nncf_graph.get_nx_graph_copy())
 
@@ -320,7 +319,6 @@ def get_ip_graph_with_merged_hw_optimized_operations(
         :param full_fusing_pattern: The GraphPatttern object representing a composition of fusing pattern variants.
         :return: The InsertionPointGraph with nodes fused according to pattern matching.
         """
-
         merged_ip_graph = deepcopy(self)
         matches = find_subgraphs_matching_pattern(merged_ip_graph.get_base_nx_graph(), full_fusing_pattern)
         for match in matches:

nncf/common/logging/track_progress.py

@@ -179,7 +179,6 @@ def __init__(
             it takes to process sequence elements. Useful when processing time is strongly non-uniform.
         :return: An iterable of the values in the sequence.
         """
-
         self.sequence = sequence
         self.weights = weights
         self.total = sum(self.weights) if self.weights is not None else total

nncf/common/pruning/mask_propagation.py

@@ -94,7 +94,6 @@ def symbolic_mask_propagation(
             are supported by the NNCF pruning algorithm.
         :return: Dict of node indices vs the decision made by symbolic mask propagation algorithm.
         """
-
         can_be_closing_convs = self._get_can_closing_convs(prunable_layers_types)
         can_prune_by_dim: Dict[int, PruningAnalysisDecision] = {k: None for k in can_be_closing_convs}  # type: ignore
         for node in self._graph.topological_sort():

nncf/common/pruning/node_selector.py

@@ -91,7 +91,6 @@ def create_pruning_groups(self, graph: NNCFGraph) -> Clusterization[NNCFNode]:
         :param graph: Graph to work with and their initialization parameters as values.
         :return: Clusterization of pruned nodes.
         """
-
         all_nodes_to_prune = graph.get_nodes_by_types(self._prune_operations_types)  # NNCFNodes here
 
         # 1. Clusters for special ops
@@ -218,7 +217,6 @@ def _pruning_dimensions_analysis(
             are supported by the NNCF pruning algorithm
         :return: Pruning node analysis after model analyzer, pruning algo compatibility and pruning dimensions checks.
         """
-
         nodes_of_group_with_non_eq_pruning_dim = self._check_internal_groups_dim(pruned_nodes_clusterization)
         can_prune_after_check_updated = can_prune_after_check.copy()
         for node_id, val in nodes_of_group_with_non_eq_pruning_dim.items():

nncf/common/quantization/initialization/range.py

@@ -87,7 +87,6 @@ def __init__(
             specified type of range initialization will be applied. It can be
             quantizers group for activations or weights.
         """
-
         super().__init__(
             range_init_config.init_type, range_init_config.num_init_samples, range_init_config.init_type_specific_params
         )

nncf/common/quantization/quantizer_propagation/graph.py

@@ -797,7 +797,6 @@ def all_outputs_are_quantized(self, node_key: str) -> bool:
         :return: True if all paths from the given node to the first
         input quantizable nodes have an activation quantizer, False otherwise.
         """
-
         nodes_keys_stack = deque(self.successors(node_key))
         while nodes_keys_stack:
             node_key = nodes_keys_stack.popleft()
@@ -1424,7 +1423,6 @@ def _handle_output_quantizers_for_weights_as_outputs_ops(
         :return: A MultiConfigQuantizerSetup with weights-as-outputs-dependent quantizers removed where possible
             and shared inputs/unified scales group adjusted to reflect the change.
         """
-
         # For the weights-are-outputs quantized operations, need to find out the dependent activation quantizers in
         # the multiconfig setup and see if it is possible to avoid requantization by selecting a common configuration
         # subset. If yes and the activation quantizer becomes unnecessary, need to unify the scales of the weight

nncf/common/quantization/quantizer_propagation/solver.py

@@ -676,7 +676,6 @@ def propagation_step(
         :param quant_prop_graph: The propagation state graph for `curr_prop_quantizer` to be propagated in.
         :return: The new state of `quant_prop_graph` with `curr_prop_quantizer` propagated one step further.
         """
-
         curr_node_key = curr_prop_quantizer.current_location_node_key
         curr_node = quant_prop_graph.nodes[curr_node_key]
         curr_node_type = curr_node[QuantizerPropagationStateGraph.NODE_TYPE_NODE_ATTR]
@@ -1002,7 +1001,6 @@ def coalesce_insertion_points(
           corresponding TargetPoints.
         :return: A list of TargetPoint groups; each group is a list of TargetPoint's.
         """
-
         if linked_scopes_groups_list is None:
             return [[ip] for ip in target_insertion_points]
         retval: List[List[TargetPoint]] = []
@@ -1313,7 +1311,6 @@ def check_transition_via_path(
           cloned before transition, which impacts the logic of the function.
         :return: The status of the transition determining how it should proceed.
         """
-
         for from_node_key, to_node_key in path:
             from_node = quant_prop_graph.nodes[from_node_key]
 
@@ -1390,7 +1387,6 @@ def get_merged_qconfigs_for_downward_branching_case(
           of the merged quantizer, if any, and the second element corresponds to configurations of the quantizers
           that would have to remain on the branches (if any).
         """
-
         if self._propagation_strategy == QuantizerPropagationRule.DO_NOT_MERGE_BRANCHES:
             # Do not merge at all
             return None, potential_qconfigs_for_each_branch

nncf/common/scopes.py

@@ -34,7 +34,6 @@ def matches_any(tested_str: str, strs_to_match_to: Union[Iterable[str], str, Non
     :return: A boolean value specifying whether a tested_str should matches at least one element
         in strs_to_match_to.
     """
-
     if strs_to_match_to is None:
         return False
 

nncf/common/utils/backend.py

@@ -136,7 +136,6 @@ def get_backend(model: Any) -> BackendType:
     :param model: The framework-specific model.
     :return: A BackendType representing the correct NNCF backend to be used when working with the framework.
     """
-
     verify_map = {
         is_torch_fx_model: BackendType.TORCH_FX,
         is_torch_model: BackendType.TORCH,

nncf/common/utils/dot_file_rw.py

@@ -85,7 +85,6 @@ def relabel_graph_for_dot_visualization(nx_graph: nx.Graph, from_reference: bool
     :param nx_graph: NetworkX graph to visualize via dot.
     :return: NetworkX graph with reserved symbols in nodes keys replaced.
     """
-
     nx_graph = copy.deepcopy(nx_graph)
 
     # .dot format reserves ':' character in node names

nncf/common/utils/patcher.py

@@ -40,7 +40,6 @@ def patch(  # noqa: C901
         :param wrapper: Wrapper function to override with.
         :param force: Whether to override previously applied patches or not.
         """
-
         obj_cls, fn_name = self.import_obj(obj_cls)
 
         # wrap only if function does exist

nncf/config/config.py

@@ -29,12 +29,14 @@
 
 @api(canonical_alias="nncf.NNCFConfig")
 class NNCFConfig(dict[str, Any]):
-    """Contains the configuration parameters required for NNCF to apply the selected algorithms.
+    """
+    Contains the configuration parameters required for NNCF to apply the selected algorithms.
 
     This is a regular dictionary object extended with some utility functions, such as the ability to attach well-defined
     structures to pass non-serializable objects as parameters. It is primarily built from a .json file, or from a
     Python JSON-like dictionary - both data types will be checked against a JSONSchema. See the definition of the
-    schema at https://openvinotoolkit.github.io/nncf/schema/, or by calling NNCFConfig.schema()."""
+    schema at https://openvinotoolkit.github.io/nncf/schema/, or by calling NNCFConfig.schema().
+    """
 
     def __init__(self, *args: Any, **kwargs: Any) -> None:
         super().__init__(*args, **kwargs)
@@ -47,7 +49,6 @@ def from_dict(cls, nncf_dict: Dict[str, Any]) -> "NNCFConfig":
 
         :param nncf_dict: A Python dict with the JSON-style configuration for NNCF.
         """
-
         cls.validate(nncf_dict)
         return cls(deepcopy(nncf_dict))
 

nncf/config/schema.py

@@ -151,8 +151,10 @@
 def validate_single_compression_algo_schema(
     single_compression_algo_dict: Dict[str, Any], ref_vs_algo_schema: Dict[str, Any]
 ) -> None:
-    """single_compression_algo_dict must conform to BASIC_COMPRESSION_ALGO_SCHEMA (and possibly has other
-    algo-specific properties"""
+    """
+    single_compression_algo_dict must conform to BASIC_COMPRESSION_ALGO_SCHEMA (and possibly has other
+    algo-specific properties
+    """
     algo_name = single_compression_algo_dict["algorithm"]
     if algo_name not in ref_vs_algo_schema:
         msg = f"Incorrect algorithm name - must be one of {str(list(ref_vs_algo_schema.keys()))}"

nncf/data/generators.py

@@ -47,7 +47,6 @@ def generate_text_data(
     :param dataset_size: Size of the data.
     :return: List of the text data ready to use.
     """
-
     try:
         import torch
     except ImportError:

nncf/experimental/common/graph/netron.py

@@ -159,7 +159,6 @@ def convert_nncf_dtype_to_ov_dtype(dtype: Dtype) -> str:
     :param dtype: The data type to be converted.
     :return: The openvino dtype string corresponding to the given data type.
     """
-
     dummy_precision_map: Dict[Dtype, str] = {Dtype.INTEGER: "i32", Dtype.FLOAT: "f32"}
 
     return dummy_precision_map[dtype]
@@ -178,7 +177,6 @@ def get_graph_desc(
     :return: A tuple containing lists of NodeDesc and EdgeDesc objects
         representing the nodes and edges of the NNCFGraph.
     """
-
     if get_attributes_fn is None:
         get_attributes_fn = lambda x: {
             "metatype": str(x.metatype.name),
@@ -257,7 +255,6 @@ def save_for_netron(
     :param get_attributes_fn: A function to retrieve additional attributes for nodes.
         Defaults to a function returning {"metatype": str(x.metatype.name)}.
     """
-
     node_descs, edge_descs = get_graph_desc(graph, include_fq_params, get_attributes_fn)
 
     net = ET.Element(Tags.NET, name=graph_name)

nncf/experimental/common/pruning/block_hierarchy.py

@@ -32,7 +32,6 @@ def __init__(self, root_groups: List[PropagationGroup]) -> None:
         :param roots: list of the root groups
         :return: networkx graph that represents the hierarchy of propagation blocks/groups.
         """
-
         self._id_counter = 0
         self._graph = nx.DiGraph()
         self._visited_block_ids_map: Dict[int, int] = {}

nncf/experimental/common/tensor_statistics/collectors.py

@@ -125,7 +125,6 @@ def __init__(
         :param window_size: Number of samples from the end of the list of collected samples to aggregate.
         Aggregates all available collected statistics in case parameter is None.
         """
-
         self._aggregation_axes = (0,) if aggregation_axes is None else aggregation_axes
         self._keepdims = True
         self._num_samples = num_samples

nncf/experimental/quantization/algorithms/post_training/pipeline.py

@@ -62,7 +62,6 @@ def experimental_create_ptq_pipeline(
         for each item of the batch or for the entire batch, default is False.
     :return: An experimental post-training quantization pipeline.
     """
-
     # Build the post-training quantization pipeline.
     pipeline_steps = []
 

nncf/experimental/tensorflow/graph/converter.py

@@ -286,7 +286,6 @@ def _collect_tfgraph_descs(graph: tf.Graph, op_names: List[str]) -> Tuple[List[N
         :param op_names: A list of names for the input operations ()
         :return: A description of nodes and edges which should be included to the NNCF graph.
         """
-
         # Traverse the `graph` and mark all ops reachable from the `input_ops`.
         # The op `u` is reachable from the `input_ops` if a directed path
         # from at least one op in `input_ops` to `u` exists in the `graph.`

nncf/experimental/torch/fx/model_transformer.py

@@ -77,7 +77,6 @@ def _traverse_graph(
         :param stop_nodes: Given stop nodes.
         :param visited: Set of already visited nodes.
         """
-
         while input_nodes:
             in_node = input_nodes.pop()
             if in_node.name in visited or in_node.name in stop_nodes:
@@ -104,7 +103,6 @@ def _apply_model_extraction(
             more than one element this function raises an assert.
         :return: Returns a submodel extracted from the given model by the given transformation.
         """
-
         transformation = transformations[-1]
         stop_nodes = set(transformation.input_node_names + transformation.output_node_names)
         visited = set()

nncf/experimental/torch/fx/quantization/quantize_pt2e.py

@@ -143,7 +143,8 @@ def quantize_pt2e(
 
 
 def _quant_node_constraint(n: torch.fx.Node) -> bool:
-    """If there is any pure ops between get_attr and quantize op they will be const propagated
+    """
+    If there is any pure ops between get_attr and quantize op they will be const propagated
     e.g. get_attr(weight) -> transpose -> quantize -> dequantize*
     (Note: dequantize op is not going to be constant propagated)
 

nncf/experimental/torch/fx/transformations.py

@@ -75,7 +75,7 @@ def _set_new_node_meta(
     model: torch.fx.GraphModule,
 ):
     """
-    Sets correct meta \"val\" value to the new node.
+    Sets correct meta 'val' value to the new node.
 
     :param new_node: The new node.
     :param prev_node: Input node of the new node.
@@ -316,7 +316,6 @@ def insert_one_qdq(model: torch.fx.GraphModule, target_point: PTTargetPoint, qua
         target node.
     :param quantizer: Quantizer module to inherit quantization parameters from.
     """
-
     # Copied from torch.ao.quantization.quantize_pt2e.convert_pt2e
     # 1. extract information for inserting q/dq node from activation_post_process
     node_type = "call_function"
@@ -613,7 +612,6 @@ def _compress_qdq_constant_transformation(model: torch.fx.GraphModule, matches)
 
     :param: model: Model to apply transformations to.
     """
-
     for match in matches:
         mul_node = match.replacements[0]
         sub_node = match.replacements[1]

nncf/experimental/torch/nas/bootstrapNAS/elasticity/elastic_width.py

@@ -294,7 +294,7 @@ def __init__(
     @property
     def width_list(self) -> List[int]:
         """
-        list of all available widths to select from. Each value corresponds to a single element in the search space of
+        List of all available widths to select from. Each value corresponds to a single element in the search space of
         operation. The search space of the model is cartesian product of search spaces of operation.
         If all widths starting from 1 to maximum number of channels with step size 1 are available, the search space
         would be prohibitively large to efficiently train and search.

nncf/experimental/torch/nas/bootstrapNAS/search/evaluator.py

@@ -216,7 +216,6 @@ def update_from_state(self, state: Dict[str, Any]) -> NoReturn:
         :param state: dict with state that should be used for updating this evaluator
         :return:
         """
-
         super().update_from_state(state)
         new_dict = state.copy()
         self._is_top1 = new_dict["is_top1"]

nncf/experimental/torch/nas/bootstrapNAS/training/training_algorithm.py

@@ -126,7 +126,6 @@ def run(
         :param tensorboard_writer: The tensorboard object to be used for logging.
         :return: the fine-tuned model and elasticity controller
         """
-
         if train_iters is None:
             train_iters = len(train_loader)
         self._training_ctrl.set_training_lr_scheduler_args(optimizer, train_iters)  # len(train_loader))

nncf/experimental/torch/search_building_blocks/search_blocks.py

@@ -304,7 +304,7 @@ def get_potential_candidate_for_block(search_graph: SearchGraph) -> Tuple[ShapeV
 
 def itemgetter_force_tuple(*indexes):
     """
-    itemgetter wrapper that always returns a tuple. The original function may return both: iterable and a single
+    Itemgetter wrapper that always returns a tuple. The original function may return both: iterable and a single
     non-iterable element, which is not convenient in the general case.
     """
     getter = itemgetter(*indexes)

nncf/experimental/torch/sparsify_activations/sparsify_activations_impl.py

@@ -225,7 +225,7 @@ def sparsify_activations(
         representing the layers to match in the model's NNCF graph; the corresponding value
         is a float number in the range [0, 1] representing the target sparsity level.
 
-        Example:
+    Example:
         ..  code-block:: python
             {
                 # Target sparsity is 60% for node "Dummy/Linear[layer]/linear_0" in the model graph
@@ -239,7 +239,6 @@ def sparsify_activations(
         filtered out internally, so there is no need to mention them in `ignored_scope`.
     :return: The sparsified model.
     """
-
     for scope, target_sparsity in target_sparsity_by_scope.items():
         if target_sparsity < 0.0 or target_sparsity > 1.0:
             msg = f'Target sparsity for scope "{scope}" should be in range [0, 1].'

nncf/experimental/torch/sparsify_activations/target_scope.py

@@ -21,7 +21,7 @@
 
 @dataclass
 class TargetScope(IgnoredScope):
-    """
+    r"""
     Specifies the target portions in a model graph.
 
     Example: