Merge pull request #92 from ziatdinovmax/master

Bump up to Numpy 2.0

Author: ziatdinovmax

.github/workflows/actions.yml

@@ -1,7 +1,7 @@
 name: build
 
 env:
-  PYTHON_MAIN_VERSION: 3.8
+  PYTHON_MAIN_VERSION: 3.10
 
 on:
   pull_request:
@@ -19,7 +19,7 @@ jobs:
     strategy:
       max-parallel: 5
       matrix:
-        python-version: [3.8, 3.9]
+        python-version: ["3.9", "3.10", "3.11"]
 
     steps:
     - uses: actions/checkout@v2

atomai/models/dgm/vae.py

@@ -434,7 +434,7 @@ def manifold2d(self, **kwargs: Union[int, List, str, bool]) -> None:  # use torc
                 figure[i * self.in_dim[0]: (i + 1) * self.in_dim[0],
                        j * self.in_dim[1]: (j + 1) * self.in_dim[1]] = imdec
         if figure.min() < 0:
-            figure = (figure - figure.min()) / figure.ptp()
+            figure = (figure - figure.min()) / np.ptp(figure)
 
         fig, ax = plt.subplots(figsize=(10, 10))
         ax.imshow(figure, cmap=cmap, origin=kwargs.get("origin", "lower"),
@@ -506,7 +506,7 @@ def manifold_traversal(self, cont_idx: int,
         grid = make_grid(torch.from_numpy(decoded),
                          nrow=d, padding=kwargs.get("pad", 2)).numpy()
         grid = grid.transpose(1, 2, 0) if len(self.in_dim) == 3 else grid[0]
-        grid = (grid - grid.min()) / grid.ptp()
+        grid = (grid - grid.min()) / np.ptp(grid)
         if not kwargs.get("keep_square", False) and disc_dim != d:
             grid = grid[:(self.in_dim[0]+kwargs.get("pad", 2)) * disc_dim]
         if plot:

atomai/models/loaders.py

@@ -34,7 +34,7 @@ def load_model(filepath: str) -> Union[Segmentor, Union[VAE, rVAE, jrVAE, jVAE],
         Model in evaluation state
     """
     device = 'cuda' if torch.cuda.is_available() else 'cpu'
-    loaded_dict = torch.load(filepath, map_location=device)
+    loaded_dict = torch.load(filepath, map_location=device, weights_only=False)
     if 'model_type' in loaded_dict.keys():
         model_type = loaded_dict.pop("model_type")
         with warnings.catch_warnings():

atomai/nets/ed.py

@@ -56,7 +56,7 @@ def __init__(self, signal_dim: Tuple[int],
         bn = kwargs.get('batch_norm', True)
         if self.downsample:
             signal_dim = [s // self.downsample for s in signal_dim]
-        n = np.product(signal_dim)
+        n = np.prod(signal_dim)
         self.reshape_ = nb_filters * n
         self.conv = ConvBlock(
             ndim, nb_layers, 1, nb_filters,
@@ -118,7 +118,7 @@ def __init__(self, signal_dim: Tuple[int],
         ndim = 2 if len(signal_dim) == 2 else 1
         if self.upsampling:
             signal_dim = [s // 4 for s in signal_dim]
-        n = np.product(signal_dim)
+        n = np.prod(signal_dim)
         self.reshape_ = (nb_filters, *signal_dim)
         self.fc = nn.Linear(z_dim, nb_filters*n)
         if self.upsampling:
@@ -269,7 +269,7 @@ def __init__(self,
         self.conv = ConvBlock(
             dim, num_layers, c, hidden_dim,
             lrelu_a=kwargs.get("lrelu_a", 0.1))
-        self.reshape_ = hidden_dim * np.product(in_dim[:2])
+        self.reshape_ = hidden_dim * np.prod(in_dim[:2])
         self.fc11 = nn.Linear(self.reshape_, latent_dim)
         self.fc12 = nn.Linear(self.reshape_, latent_dim)
         self._out = nn.Softplus() if kwargs.get("softplus_out") else lambda x: x
@@ -323,7 +323,7 @@ def __init__(self,
         super(fcEncoderNet, self).__init__()
         dense = []
         for i in range(num_layers):
-            input_dim = np.product(in_dim) if i == 0 else hidden_dim
+            input_dim = np.prod(in_dim) if i == 0 else hidden_dim
             dense.extend([nn.Linear(input_dim, hidden_dim), nn.Tanh()])
         self.dense = nn.Sequential(*dense)
         self.reshape_ = hidden_dim
@@ -335,7 +335,7 @@ def forward(self, x: torch.Tensor):
         """
         Forward pass
         """
-        x = x.reshape(-1, np.product(x.size()[1:]))
+        x = x.reshape(-1, np.prod(x.size()[1:]))
         x = self.dense(x)
         x = x.reshape(-1, self.reshape_)
         z_mu = self.fc11(x)
@@ -378,7 +378,7 @@ def __init__(self,
         super(jfcEncoderNet, self).__init__()
         dense = []
         for i in range(num_layers):
-            input_dim = np.product(in_dim) if i == 0 else hidden_dim
+            input_dim = np.prod(in_dim) if i == 0 else hidden_dim
             dense.extend([nn.Linear(input_dim, hidden_dim), nn.Tanh()])
         self.dense = nn.Sequential(*dense)
         self.reshape_ = hidden_dim
@@ -394,7 +394,7 @@ def forward(self, x: torch.Tensor):
         """
         Forward pass
         """
-        x = x.reshape(-1, np.product(x.size()[1:]))
+        x = x.reshape(-1, np.prod(x.size()[1:]))
         x = self.dense(x)
         x = x.reshape(-1, self.reshape_)
         encoded = [self.fc11(x), self._out(self.fc12(x))]
@@ -446,7 +446,7 @@ def __init__(self,
         self.conv = ConvBlock(
             dim, num_layers, c, hidden_dim,
             lrelu_a=kwargs.get("lrelu_a", 0.1))
-        self.reshape_ = hidden_dim * np.product(in_dim[:2])
+        self.reshape_ = hidden_dim * np.prod(in_dim[:2])
         self.fc11 = nn.Linear(self.reshape_, latent_dim)
         self.fc12 = nn.Linear(self.reshape_, latent_dim)
         fc13 = []
@@ -501,7 +501,7 @@ def __init__(self,
         dim = 2 if len(out_dim) > 1 else 1
         c = out_dim[-1] if len(out_dim) > 2 else 1
         self.fc_linear = nn.Linear(
-            latent_dim, hidden_dim * np.product(out_dim[:2]),
+            latent_dim, hidden_dim * np.prod(out_dim[:2]),
             bias=False)
         self.reshape_ = (hidden_dim, *out_dim[:2])
         self.decoder = ConvBlock(
@@ -563,7 +563,7 @@ def __init__(self,
             hidden_dim_ = latent_dim if i == 0 else hidden_dim
             decoder.extend([nn.Linear(hidden_dim_, hidden_dim), nn.Tanh()])
         self.decoder = nn.Sequential(*decoder)
-        self.out = nn.Linear(hidden_dim, np.product(out_dim))
+        self.out = nn.Linear(hidden_dim, np.prod(out_dim))
         self.out_dim = (c, *out_dim[:2])
 
     def forward(self, z: torch.Tensor) -> torch.Tensor:

atomai/predictors/predictor.py

@@ -286,7 +286,8 @@ def run(self,
             return decoded_imgs
         images, decoded_imgs = self.predict(
             image_data, return_image=True, **kwargs)
-        loc = Locator(self.thresh, refine=self.refine, d=self.d)
+        thresh = kwargs.get("thresh", self.thresh)
+        loc = Locator(thresh, refine=self.refine, d=self.d)
         coordinates = loc.run(decoded_imgs, images)
         if self.verbose:
             n_images_str = " image was " if decoded_imgs.shape[0] == 1 else " images were "

atomai/transforms/imaug.py

@@ -319,7 +319,7 @@ def run(self, images: np.ndarray, targets: np.ndarray) -> Tuple[np.ndarray]:
             pass
         else:
             raise NotImplementedError("Use 'channel_first' or 'channel_last'")
-        images = (images - images.min()) / images.ptp()
+        images = (images - images.min()) / np.ptp(images)
         if self.custom_transform is not None:
             images, targets = self.custom_transform(images, targets)
         if self.rotation and same_dim:
@@ -354,7 +354,7 @@ def run(self, images: np.ndarray, targets: np.ndarray) -> Tuple[np.ndarray]:
             images = np.expand_dims(images, axis=3)
         else:
             raise NotImplementedError("Use 'channel_first' or 'channel_last'")
-        images = (images - images.min()) / images.ptp()
+        images = (images - images.min()) / np.ptp(images)
         return images, targets
 
 

atomai/utils/preproc.py

@@ -757,7 +757,7 @@ def torch_format_image(image_data: np.ndarray,
     else:
         pass
     if norm:
-        image_data = (image_data - image_data.min()) / image_data.ptp()
+        image_data = (image_data - image_data.min()) / np.ptp(image_data)
     image_data = torch.from_numpy(image_data).float()
     return image_data
 
@@ -786,7 +786,7 @@ def torch_format_spectra(spectra: np.ndarray,
     else:
         pass
     if norm:
-        spectra = (spectra - spectra.min()) / spectra.ptp()
+        spectra = (spectra - spectra.min()) / np.ptp(spectra)
     spectra = torch.from_numpy(spectra).float()
     return spectra
 

requirements.txt

@@ -12,4 +12,4 @@ opencv-python>=4.1.0
 networkx>=2.5
 mendeleev<=0.6.1
 progressbar2>=3.38.0
-gpytorch>=1.4.0,<1.7.0
+gpytorch>=1.9.1