balance weights:

avoid using dacapo specific classes in utils. Utils is a place to put
code that is common and should maybe be provided by a seperate library

dacapo/experiments/tasks/predictors/distance_predictor.py

@@ -12,6 +12,7 @@
 import torch
 
 import logging
+from typing import List
 
 logger = logging.getLogger(__name__)
 
@@ -76,13 +77,16 @@ def create_weight(self, gt, target, mask):
             )
         else:
             distance_mask = np.ones_like(target.data)
-        return balance_weights(
-            gt,
-            2,
-            slab=tuple(1 if c == "c" else -1 for c in gt.axes),
-            masks=[
-                NumpyArray.from_np_array(distance_mask, gt.roi, gt.voxel_size, gt.axes)
-            ],
+        return NumpyArray.from_np_array(
+            balance_weights(
+                gt[target.roi],
+                2,
+                slab=tuple(1 if c == "c" else -1 for c in gt.axes),
+                masks=[distance_mask],
+            ),
+            gt.roi,
+            gt.voxel_size,
+            gt.axes,
         )
 
     @property

dacapo/utils/balance_weights.py

@@ -3,35 +3,34 @@
 import numpy as np
 
 import itertools
+from typing import List
 
 
 def balance_weights(
-    labels,
-    num_classes,
-    masks=list(),
+    label_data: np.ndarray,
+    num_classes: int,
+    masks: List[np.ndarray] = list(),
     slab=None,
-    clipmin=0.05,
-    clipmax=0.95,
+    clipmin: float = 0.05,
+    clipmax: float = 0.95,
 ):
-
-    label_data = labels[labels.roi]
-
-    assert len(np.unique(label_data)) <= num_classes, (
-        "Found more unique labels than classes in %s." % labels
-    )
-    assert 0 <= np.min(label_data) < num_classes, (
-        "Labels %s are not in [0, num_classes)." % labels
-    )
-    assert 0 <= np.max(label_data) < num_classes, (
-        "Labels %s are not in [0, num_classes)." % labels
-    )
+    unique_labels = np.unique(label_data)
+    assert (
+        len(unique_labels) <= num_classes
+    ), f"Found unique labels {unique_labels} but expected only {num_classes}."
+    assert (
+        0 <= np.min(label_data) < num_classes
+    ), f"Labels {unique_labels} are not in [0, {num_classes})."
+    assert (
+        0 <= np.max(label_data) < num_classes
+    ), f"Labels {unique_labels} are not in [0, {num_classes})."
 
     # initialize error scale with 1s
     error_scale = np.ones(label_data.shape, dtype=np.float32)
 
     # set error_scale to 0 in masked-out areas
     for mask in masks:
-        error_scale *= mask[labels.roi]
+        error_scale *= mask
 
     if slab is None:
         slab = error_scale.shape
@@ -71,7 +70,4 @@ def balance_weights(
         # scale_slab the masked-in scale_slab with the class weights
         scale_slab *= np.take(w, labels_slab)
 
-    weights = NumpyArray.from_np_array(
-        error_scale, labels.roi, labels.voxel_size, labels.axes
-    )
-    return weights
+    return error_scale