Run inference with pytorch 1.11 and numpy > 1.20

echofilter/data/transforms.py

@@ -129,7 +129,6 @@ def __init__(self, output_size, order=1):
         self.order = order
 
     def __call__(self, sample):
-
         order = self.order
         if order is None:
             order = np.random.randint(1)
@@ -148,7 +147,7 @@ def __call__(self, sample):
                     "ignore", "Bi-quadratic interpolation behavior has changed"
                 )
                 sample[key] = skimage.transform.resize(
-                    np.asarray(sample[key]).astype(np.float),
+                    np.asarray(sample[key]).astype(float),
                     self.output_size,
                     order=order,
                     clip=False,
@@ -224,7 +223,6 @@ def __init__(self, *args, p=0.5, **kwargs):
         self.p = p
 
     def __call__(self, sample):
-
         if random.random() > self.p:
             # Nothing to do
             return sample
@@ -341,7 +339,6 @@ def __init__(self, output_size, p=0.5, sigma=8.0, alpha=0.05, order=1):
         self.alpha = _pair(alpha)
 
     def __call__(self, sample):
-
         if random.random() > self.p:
             # Nothing to do
             return sample
@@ -464,7 +461,6 @@ def __init__(self, center, deviation, robust2stdev=True):
         self.robust2stdev = robust2stdev
 
     def __call__(self, sample):
-
         if not isinstance(self.center, str):
             center = self.center
         elif self.center.lower() == "mean":
@@ -521,7 +517,6 @@ def __init__(self, nan_val=0.0):
         self.nan_val = nan_val
 
     def __call__(self, sample):
-
         # Can't use np.nan_to_num to assign nan to a specific value if
         # numpy version <= 1.17.
         sample["signals"][np.isnan(sample["signals"])] = self.nan_val
@@ -546,7 +541,6 @@ def __init__(self, axis=0, p=0.5):
         self.p = p
 
     def __call__(self, sample):
-
         if random.random() > self.p:
             # Nothing to do
             return sample
@@ -575,7 +569,6 @@ def __init__(self, max_crop_fraction):
         self.max_crop_fraction = max_crop_fraction
 
     def __call__(self, sample):
-
         width = sample["signals"].shape[0]
 
         crop_fraction = random.uniform(0.0, self.max_crop_fraction)
@@ -716,7 +709,6 @@ def __init__(
         self.fraction_close = fraction_close
 
     def __call__(self, sample):
-
         p = random.random()
 
         # Possibility of doing no crop at all

echofilter/nn/modules/pathing.py

@@ -67,7 +67,12 @@ def forward(self, x1, x2):
         if diffX != 0 or diffY != 0:
             x1 = F.pad(
                 x1,
-                [diffX // 2, diffX - diffX // 2, diffY // 2, diffY - diffY // 2],
+                [
+                    torch.div(diffX, 2, rounding_mode="trunc"),
+                    diffX - torch.div(diffX, 2, rounding_mode="trunc"),
+                    torch.div(diffY, 2, rounding_mode="trunc"),
+                    diffY - torch.div(diffY, 2, rounding_mode="trunc"),
+                ],
                 mode="replicate",
             )
         # if you have padding issues, see

echofilter/raw/loader.py

@@ -193,7 +193,7 @@ def transect_loader(
     timestamps = np.empty((n_lines - skip_lines))
     timestamps[:] = np.nan
 
-    row_lengths = np.empty((n_lines - skip_lines), dtype=np.int)
+    row_lengths = np.empty((n_lines - skip_lines), dtype=int)
     row_depth_starts = np.empty((n_lines - skip_lines))
     row_depth_ends = np.empty((n_lines - skip_lines))