Add functions to count classes appearing in prediction sets

Author: mariya

src/conformist/validation_run.py

@@ -117,6 +117,25 @@ def mean_fnrs_by_class(self, sets, class_names):
             averages[key] = sum(sizes) / len(sizes)
         return averages
 
+    def prediction_counts_by_class(self, class_names, coocurring_only=False):
+        counts = {}
+        for i in range(len(self.prediction_sets)):
+            labels = self.prediction_sets[i]
+
+            # Get corresponding values from class_names
+            pset_class_names = [class_names[i] for i, label in enumerate(labels) if label == 1]
+
+            do_count = True
+            if coocurring_only:
+                do_count = len(pset_class_names) > 1
+
+            if do_count:
+                for class_name in pset_class_names:
+                    class_counts = counts.get(class_name, 0)
+                    class_counts += 1
+                    counts[class_name] = class_counts
+        return counts
+
     def run_reports(self, base_output_dir):
         mean_set_sizes = self.mean_set_sizes_by_class(self.class_names)
         mean_fnrs = self.mean_fnrs_by_class(self.prediction_sets, self.class_names)

src/conformist/validation_trial.py

@@ -115,6 +115,24 @@ def mean_fnrs_by_class(self, class_names):
                 means[class_name] = statistics.mean(d_means)
         return means
 
+    def mean_prediction_counts_by_class(self, class_names, coocurring_only=False):
+        prediction_count_dicts = []
+        for run in self.runs:
+            prediction_count_dicts.append(run.prediction_counts_by_class(class_names, coocurring_only))
+
+        means = {}
+        for class_name in class_names:
+            d_means = []
+            for d in prediction_count_dicts:
+                if class_name in d:
+                    d_means.append(d[class_name])
+            if len(d_means) > 0:
+                means[class_name] = statistics.mean(d_means)
+
+        # Sort by count descending
+        means = dict(sorted(means.items(), key=lambda item: item[1], reverse=True))
+        return means
+
     def run_reports(self, base_output_dir):
         self.create_output_dir(base_output_dir)
         self.visualize_empirical_fnr()