update compute

charmatch.py

@@ -87,25 +87,34 @@ class charmatch(evaluate.Metric):
         # TODO: Download external resources if needed
         pass
 
-    def _compute(self, inputs, expected, outputs):
+    def _compute(inputs, expected, outputs):
         def get_score(input, expected, output):
             print(input, expected, output)
-            expected_corrections = lev(input, expected)
-            distance_to_input = lev(input, output)
-            distance_to_expected = lev(output, expected)
-            print(f'dl(s,g): {expected_corrections}\ndl(s,h): {distance_to_input}\ndl(h,g): {distance_to_expected}')
-
-            true_positives = min(expected_corrections, max(0, (expected_corrections + distance_to_input - distance_to_expected))) / 2
-            print(f'T: {true_positives}')
-
-            precision = true_positives / distance_to_input
-            recall = true_positives / expected_corrections
-            f_05 = (1 + 0.5**2) * (precision * recall) / (0.5**2 * precision + recall)
-            print(f'P: {precision}\nR: {recall}')
- 
-            return f_05
-
-        avg = sum([get_score(*row) for row in zip(inputs, expected, outputs)]) / len(inputs)
+            deduped = {input, expected, output}
+            if len(deduped) == 1:
+                return 1.0
+            elif len(deduped) == 2:
+                if expected == output:
+                    return 1.0
+                else:
+                    return 0.0
+            else:
+                expected_corrections = lev(input, expected)
+                distance_to_input = lev(input, output)
+                distance_to_expected = lev(output, expected)
+                print(f'dl(s,g): {expected_corrections}\ndl(s,h): {distance_to_input}\ndl(h,g): {distance_to_expected}')
+
+                true_positives = min(expected_corrections, max(0, (expected_corrections + distance_to_input - distance_to_expected))) / 2
+                print(f'T: {true_positives}')
+
+                precision = true_positives / distance_to_input
+                recall = true_positives / expected_corrections
+                f_05 = (1 + 0.5**2) * (precision * recall) / (0.5**2 * precision + recall)
+                print(f'P: {precision}\nR: {recall}')
+
+                return f_05
+
+        avg = sum([get_score(*row) for row in zip(inputs, expected, outputs)]) / len(inputs) * 100
 
         return {
             "fscore": avg