remove numpy.typing

Author: AlexanderDokuchaev

.github/workflows/weekly.yml

@@ -14,7 +14,7 @@ jobs:
       matrix:
         include:
           - python_version: "3.9"
-            override_requirements: "numpy==1.19.0"
+            override_requirements: "numpy==1.19.1"
           - python_version: "3.9"
             override_requirements: "numpy==1.20.0"
           - python_version: "3.9"

nncf/common/tensor_statistics/collectors.py

@@ -15,7 +15,6 @@
 from typing import Any, Deque, Dict, List, Optional, Tuple, Union, cast
 
 import numpy as np
-from numpy.typing import NDArray
 
 from nncf.common.tensor import NNCFTensor
 from nncf.common.tensor import TensorType
@@ -318,7 +317,7 @@ class MedianMADStatisticCollector(OfflineTensorStatisticCollector):
     Collector estimates median and median absolute deviation (MAD).
     """
 
-    def _prepare_statistics(self) -> Tuple[NDArray[Any], NDArray[Any]]:
+    def _prepare_statistics(self) -> Tuple[np.ndarray, np.ndarray]:
         per_channel_history = get_per_channel_history(
             self._samples, cast(List[int], self._reduction_shape), discard_zeros=True
         )

nncf/common/tensor_statistics/reduction.py

@@ -12,7 +12,6 @@
 from typing import Any, Deque, List, Tuple
 
 import numpy as np
-from numpy.typing import NDArray
 
 
 def get_channel_count_and_dim_idx(scale_shape: List[int]) -> Tuple[int, int]:
@@ -25,7 +24,7 @@ def get_channel_count_and_dim_idx(scale_shape: List[int]) -> Tuple[int, int]:
     return channel_count, channel_dim_idx
 
 
-def split_into_channels(input_: NDArray[Any], scale_shape: List[int]) -> List[NDArray[Any]]:
+def split_into_channels(input_: np.ndarray, scale_shape: List[int]) -> List[np.ndarray]:
     channel_count, channel_dim_idx = get_channel_count_and_dim_idx(scale_shape)
     channel_first_tensor = np.moveaxis(input_, channel_dim_idx, 0)
     if channel_count == 1:
@@ -61,7 +60,7 @@ def get_per_channel_history(
     return per_channel_history
 
 
-def np_percentile_reduce_like(input_: NDArray[Any], ref_tensor_shape: Tuple[int], q: float) -> NDArray[Any]:
+def np_percentile_reduce_like(input_: np.ndarray, ref_tensor_shape: Tuple[int], q: float) -> np.ndarray:
     numel = np.prod(ref_tensor_shape)
     if numel == 1:
         return np.array([np.percentile(input_, q)])