⬆️ Upgrade numpy Dependency (#543)

- np.int, np.bool, np.float is depreciated in v1.20 in favour of np.int_, np.bool_, np.float_

Waiting on some dependencies to update:
- numba not yet compatible with numpy 1.24
  >   from numba.np.ufunc import _internal
  E   SystemError: initialization of _internal failed without raising an exception

   - numba/numba#8464
   - numba/numba#8691
   - numba/numba#8841
   - https://github.com/numba/numba/milestone/63

- [x] Waiting for next numba release with numpy 1.24 support.

---------
Co-authored-by: John Pocock <[email protected]>

Author: shaneahmed

requirements/requirements.txt

@@ -9,7 +9,8 @@ imagecodecs>=2022.9.26
 joblib>=1.1.1
 jupyterlab>=3.5.2
 matplotlib>=3.6.2
-numpy>=1.23.5, <1.24 # v1.24 produces error on Windows
+numba>=0.57.0
+numpy>=1.23.5
 opencv-python>=4.6.0
 openslide-python>=1.2.0
 pandas>=2.0.0

tests/models/test_patch_predictor.py

@@ -134,7 +134,9 @@ def test_patch_dataset_crash(tmp_path):
         _ = PatchDataset(imgs)
 
     # ndarray of mixed dtype
-    imgs = np.array([np.random.randint(0, 255, (4, 5, 3)), "Should crash"])
+    imgs = np.array(
+        [np.random.randint(0, 255, (4, 5, 3)), "Should crash"], dtype=object
+    )
     with pytest.raises(ValueError, match="Provided input array is non-numerical."):
         _ = PatchDataset(imgs)
 

tests/test_utils.py

@@ -52,7 +52,7 @@ def test_imresize():
     # test for dtype conversion, pairs of
     # (original type, converted type)
     test_dtypes = [
-        (np.bool, np.uint8),
+        (np.bool_, np.uint8),
         (np.int8, np.int16),
         (np.int16, np.int16),
         (np.int32, np.float32),
@@ -1053,8 +1053,8 @@ def edge_mask(bounds: Tuple[int, int, int, int]) -> np.ndarray:
         l, t, r, b = bounds
         slide_width, slide_height = slide_dimensions
         x, y = np.meshgrid(np.arange(l, r), np.arange(t, b), indexing="ij")
-        under = np.logical_or(x < 0, y < 0).astype(np.int)
-        over = np.logical_or(x >= slide_width, y >= slide_height).astype(np.int)
+        under = np.logical_or(x < 0, y < 0).astype(np.int_)
+        over = np.logical_or(x >= slide_width, y >= slide_height).astype(np.int_)
         return under, over
 
     loc = (-5, -5)

tests/test_wsireader.py

@@ -186,7 +186,7 @@ def read_bounds_level_consistency(wsi, bounds):
     # from interpolation when calculating the downsampled levels. This
     # adds some tolerance for the comparison.
     blurred = [cv2.GaussianBlur(img, (5, 5), cv2.BORDER_REFLECT) for img in resized]
-    as_float = [img.astype(np.float) for img in blurred]
+    as_float = [img.astype(np.float_) for img in blurred]
 
     # Pair-wise check resolutions for mean squared error
     for i, a in enumerate(as_float):
@@ -2373,7 +2373,7 @@ def test_read_rect_level_consistency(sample_key, reader_class, kwargs):
         # from interpolation when calculating the downsampled levels. This
         # adds some tolerance for the comparison.
         blurred = [cv2.GaussianBlur(img, (5, 5), cv2.BORDER_REFLECT) for img in resized]
-        as_float = [img.astype(np.float) for img in blurred]
+        as_float = [img.astype(np.float_) for img in blurred]
 
         # Pair-wise check resolutions for mean squared error
         for i, a in enumerate(as_float):

tiatoolbox/tools/registration/wsi_registration.py

@@ -486,7 +486,7 @@ def compute_feature_distances(
             axis=len(features_x.shape),
         )
 
-        feature_size_2d = np.int(np.sqrt(feature_distance.shape[0]))
+        feature_size_2d = np.int_(np.sqrt(feature_distance.shape[0]))
         ref_feature_size_2d = factor * feature_size_2d
         feature_size, ref_feature_size = feature_size_2d**2, ref_feature_size_2d**2
         feature_grid = np.kron(

tiatoolbox/utils/metrics.py

@@ -90,8 +90,8 @@ def dice(gt_mask, pred_mask):
     if gt_mask.shape != pred_mask.shape:
         raise ValueError(f'{"Shape mismatch between the two masks."}')
 
-    gt_mask = gt_mask.astype(np.bool)
-    pred_mask = pred_mask.astype(np.bool)
+    gt_mask = gt_mask.astype(np.bool_)
+    pred_mask = pred_mask.astype(np.bool_)
     sum_masks = gt_mask.sum() + pred_mask.sum()
     if sum_masks == 0:
         return np.NAN

tiatoolbox/utils/transforms.py

@@ -112,7 +112,7 @@ def imresize(img, scale_factor=None, output_size=None, interpolation="optimise")
     # error). The `converted type` has been selected so that
     # they can maintain the numeric precision of the `original type`.
     dtype_mapping = [
-        (np.bool, np.uint8),
+        (np.bool_, np.uint8),
         (np.int8, np.int16),
         (np.int16, np.int16),
         (np.int32, np.float32),

tiatoolbox/wsicore/wsireader.py

@@ -911,15 +911,15 @@ def _find_tile_params(
             level = np.log2(rescale)
             if not level.is_integer():
                 raise ValueError
-            level = np.int(level)
+            level = np.int_(level)
             slide_dimension = self.info.level_dimensions[level]
             rescale = 1
         # Raise index error if desired pyramid level not embedded
         # in level_dimensions
         except IndexError:
             level = 0
             slide_dimension = self.info.level_dimensions[level]
-            rescale = np.int(rescale)
+            rescale = np.int_(rescale)
             logger.warning(
                 "Reading WSI at level 0. Desired tile_objective_value %s "
                 "not available.",
@@ -2508,7 +2508,7 @@ def _info(self):
         box = glymur_wsi.box
         description = box[3].xml.find("description")
         matches = re.search(r"(?<=AppMag = )\d\d", description.text)
-        objective_power = np.int(matches[0])
+        objective_power = np.int_(matches[0])
         image_header = box[2].box[0]
         slide_dimensions = (image_header.width, image_header.height)