bugfixes

vanvalenlab · Aug 28, 2020 · 71d5dff · 71d5dff
1 parent cb17a10
commit 71d5dff
Show file tree

Hide file tree

Showing 4 changed files with 7 additions and 5 deletions.
diff --git a/caliban_toolbox/build.py b/caliban_toolbox/build.py
@@ -170,7 +170,8 @@ def pad_image_stack(images, crop_size):
         # don't need to pad
         return images
     else:
-        new_images = np.zeros((images.shape[0], new_row_len, new_col_len, images.shape[3]))
+        new_images = np.zeros((images.shape[0], new_row_len, new_col_len, images.shape[3]),
+                              dtype=images.dtype)
         new_images[:, :row_len, :col_len, :] = images
         return new_images
 

diff --git a/caliban_toolbox/dataset_benchmarker.py b/caliban_toolbox/dataset_benchmarker.py
@@ -86,7 +86,7 @@ def _benchmark_category(self, category_ids):
         stats_dict = {}
         for cat in unique_categories:
             # cat the index of metrics corresponding to current category
-            cat_idx = np.isin(unique_categories, cat)
+            cat_idx = np.isin(category_ids, cat)
             cat_dict = {}
 
             # sum metrics across individual images within current category
@@ -96,7 +96,7 @@ def _benchmark_category(self, category_ids):
             # compute additional metrics not produced by Metrics class
             cat_dict['recall'] = cat_dict['correct_detections'] / cat_dict['n_true']
 
-            cat_dict['precision'] = cat_dict['correct_detections'] / stats_dict['n_pred']
+            cat_dict['precision'] = cat_dict['correct_detections'] / cat_dict['n_pred']
 
             cat_dict['f1'] = hmean([cat_dict['recall'], cat_dict['precision']])
 

diff --git a/caliban_toolbox/dataset_builder.py b/caliban_toolbox/dataset_builder.py
@@ -535,6 +535,7 @@ def build_dataset(self, tissues='all', platforms='all', output_shape=(512, 512),
             current_dict = self._clean_labels(data_dict=current_dict, relabel=relabel,
                                               small_object_threshold=small_object_threshold,
                                               min_objects=min_objects)
+            print("index is {}, unique is {}".format(idx, np.unique(current_dict['y'])))
             dicts[idx] = current_dict
 
         return dicts

diff --git a/caliban_toolbox/dataset_builder_test.py b/caliban_toolbox/dataset_builder_test.py
@@ -36,7 +36,7 @@
 
 def _create_test_npz(path, constant_value=1, X_shape=(10, 20, 20, 3), y_shape=(10, 20, 20, 1)):
     X_data = np.full(X_shape, constant_value)
-    y_data = np.full(y_shape, constant_value, dtype='int16')
+    y_data = np.full(y_shape, constant_value * 2, dtype='int16')
     np.savez(path, X=X_data, y=y_data)
 
 
@@ -553,7 +553,7 @@ def test_build_dataset(tmp_path):
         assert X_base.shape[0] * 4 == X_crop.shape[0]
 
     # check that NPZs have been relabeled
-    for current_dict in output_dicts:
+    for current_dict in output_dicts_crop:
         assert len(np.unique(current_dict['y'])) == 2
 
     # different sizes for different splits