Merge pull request #88 from clamsproject/87-fix-neg-score

TimePoint outputs now include softmax for `NEG` label

app.py

@@ -117,9 +117,8 @@ def _annotate(self, mmif: Union[str, dict, Mmif], **parameters) -> Mmif:
         if self.logger.isEnabledFor(logging.DEBUG):
             self.logger.debug(f"Processing took {time.perf_counter() - t} seconds")
 
-        tp_labelset = FRAME_TYPES
         new_view.new_contain(AnnotationTypes.TimePoint, 
-                             document=vd.id, timeUnit='milliseconds', labelset=tp_labelset)
+                             document=vd.id, timeUnit='milliseconds', labelset=self.classifier.postbin_labels)
 
         for prediction in predictions:
             timepoint_annotation = new_view.new_annotation(AnnotationTypes.TimePoint)

modeling/stitch.py

@@ -61,7 +61,7 @@ def collect_timeframes(self, predictions: list) -> list:
         if self.debug:
             print('>>> labels', labels)
         timeframes = []
-        open_frames = { label: TimeFrame(label, self) for label in labels}
+        open_frames = {label: TimeFrame(label, self) for label in labels}
         for prediction in predictions:
             if self.debug:
                 print(prediction)

modeling/train.py

@@ -265,17 +265,19 @@ def pre_bin_label_names(config, raw_labels=None):
     if 'pre' in config["bins"]:
         return list(config["bins"]["pre"].keys()) + [modeling.negative_label]
     elif raw_labels is not None:
-        return raw_labels
+        return raw_labels + [modeling.negative_label]
     else:
         return []
 
+
 def post_bin_label_names(config):
     post_labels = list(config["bins"].get("post", {}).keys())
     if post_labels:
         return post_labels + [modeling.negative_label]
     else:
         return pre_bin_label_names(config)
 
+
 def get_final_label_names(config):
     if config and "post" in config["bins"]:
         return post_bin_label_names(config)
@@ -346,7 +348,6 @@ def train_model(model, loss_fn, device, train_loader, valid_loader, configs, n_l
 
 def export_train_result(out: IO, predictions: Tensor, labels: Tensor, labelset: List[str], img_enc_name: str):
     """Exports the data into a human-readable format.
-    @return: class-based accuracy metrics for each label, organized into a csv.
     """
 
     label_metrics = defaultdict(dict)