Revert "tf-unet-fix" and fix. (#237)

Author: jan-grzybek-ampere

utils/cv/brats.py

@@ -236,10 +236,12 @@ def submit_predictions(self, prediction):
         )
         self.__current_img_id += 1
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         from utils.cv.nnUNet.nnunet.evaluation.region_based_evaluation import evaluate_regions, get_brats_regions
         evaluate_regions(
-            self.__processed_predictions_dir_path, Path(self.__preprocessed_dir_path, "labelsTr"), get_brats_regions()
+            str(self.__processed_predictions_dir_path),
+            str(Path(self.__preprocessed_dir_path, "labelsTr")),
+            get_brats_regions()
         )
         with open(Path(self.__processed_predictions_dir_path, "summary.csv")) as f:
             for line in f:

utils/cv/coco.py

@@ -171,13 +171,13 @@ def submit_mask_prediction(self, id_in_batch, bbox, score, category, mask):
             "segmentation": mask
         })
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         """
         A function summarizing the accuracy achieved on the images obtained with get_input_array() calls on which
         predictions done where supplied with submit_bbox_prediction() function.
         """
         if self.do_skip():
-            return {}
+            return
 
         if self._task == "bbox":
             predictions = np.array(self._predictions)

utils/cv/imagenet.py

@@ -152,13 +152,13 @@ def submit_predictions(self, id_in_batch: int, top_1_index: int, top_5_indices:
         self.__top_1_count += int(ground_truth == top_1_index)
         self.__top_5_count += int(ground_truth in top_5_indices)
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         """
         A function summarizing the accuracy achieved on the images obtained with get_input_array() calls on which
         predictions done where supplied with submit_predictions() function.
         """
         if self.do_skip():
-            return {}
+            return
 
         top_1_accuracy = self.__top_1_count / self.__current_img
         # print("\n Top-1 accuracy = {:.3f}".format(top_1_accuracy))

utils/cv/kits.py

@@ -232,13 +232,13 @@ def submit_predictions(self, prediction):
             self.__calc_dice_score(full_prediction, ground_truth)
             self.__current_img_id += 1
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         if self.__current_img_id < 1:
             utils.print_warning_message(
                 "Not a single image has been completed - cannot calculate accuracy. Note that images of KiTS dataset "
                 "are processed in slices due to their size. That implies that complete processing of one image can "
                 "involve many passes through the network.")
-            return {"mean_kidney_acc": None, "mean_tumor_acc": None, "mean_composite_acc": None}
+            return
 
         mean_kidney = self.__kidney_score / self.__current_img_id
         # print("\n Mean kidney segmentation accuracy = {:.3f}".format(mean_kidney))

utils/helpers.py

@@ -48,19 +48,32 @@ def do_skip(self) -> bool:
     def reset(self) -> bool:
         raise NotImplementedError
 
-    def summarize_accuracy(self) -> dict:
+    def _summarize_accuracy(self) -> dict:
         raise NotImplementedError
 
+    # don't override this method, override _summarize_accuracy instead
+    def summarize_accuracy(self) -> dict:
+        accuracy_results = self._summarize_accuracy()
+        if accuracy_results is None:
+            accuracy_results = {}
+        assert type(accuracy_results) is dict
+        for k, v in accuracy_results.items():
+            assert isinstance(k, str)
+            try:
+                float(v)
+            except Exception as e:
+                raise e
+        return accuracy_results
+
     def print_accuracy_metrics(self) -> dict:
         accuracy_results = self.summarize_accuracy()
-        assert type(accuracy_results) is dict
-        if len(accuracy_results) == 0 or None in accuracy_results.values():
-            print_warning_message("Accuracy metrics are unavailable.")
+        if len(accuracy_results) == 0:
+            print_warning_message("No accuracy metrics to print.")
         else:
             max_len = 14
             indent = 2 * " "
             print(f"\n{indent}ACCURACY")
             for metric in accuracy_results.keys():
-                print(f"{3 * indent}{metric}{(max_len - len(metric)) * ' '}{3 * indent}" +
+                print(f"{3 * indent}{metric[:max_len]}{(max_len - len(metric)) * ' '}{3 * indent}" +
                       "= {:>7.3f}".format(float(accuracy_results[metric])))
         return accuracy_results

utils/nlp/alpaca_instruct.py

@@ -102,7 +102,7 @@ def metric_max_over_ground_truth(metric_fn, pred, gt):
         self._f1 += metric_max_over_ground_truth(f1_score, answer, ground_truth)
         self._count += 1
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         exact_match = self._exact_match / self._count
         f1 = self._f1 / self._count
         return {"exact_match": exact_match, "f1": f1}

utils/nlp/cnn_dailymail.py

@@ -212,7 +212,7 @@ def get_bigrams(text):
         self.__rouge2_count += rouge2_score(normalize(summary), normalize(ground_truth))
         self.__unsubmitted_count -= 1
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         """
         A function summarizing the accuracy achieved on the texts obtained with get_*_array() calls on which
         predicted summaries were supplied with submit_predictions() function.

utils/nlp/conll2003.py

@@ -183,7 +183,7 @@ def metric_max_over_ground_truths(metric_fn, prediction, ground_truth):
         self.__f1_count += metric_max_over_ground_truths(f1_score, prediction, ground_truth)
         self.__unsubmitted_count -= 1
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         """
         A function summarizing the accuracy achieved on the sequences obtained with get_*_array() calls on which
         predictions were supplied with submit_predictions() function.

utils/nlp/lambada.py

@@ -156,7 +156,7 @@ def metric_max_over_ground_truths(metric_fn, prediction, ground_truths):
         self.__f1_count += metric_max_over_ground_truths(f1_score, answer, ground_truths)
         self.__unsubmitted_count -= 1
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         """
         A function summarizing the accuracy achieved on the questions obtained with get_*_array() calls on which
         predicted answers were supplied with submit_predictions() function.

utils/nlp/mrpc.py

@@ -83,7 +83,7 @@ def submit_predictions(self, prediction, label):
         if label == prediction:
             self.__correct += 1
 
-    def summarize_accuracy(self):
+    def _summarize_accuracy(self):
         """
         A function summarizing the obtained accuracy for the model