update compute

charmatch.py

@@ -71,9 +71,9 @@ class charmatch(evaluate.Metric):
             inputs_description="input expected output",
             # This defines the format of each prediction and reference
             features=datasets.Features({
-                'inputs': datasets.Sequence(datasets.Value('string')),
-                'expected': datasets.Sequence(datasets.Value('string')),
-                'outputs': datasets.Sequence(datasets.Value('string'))
+                'input': datasets.Value('string'),
+                'expected': datasets.Value('string'),
+                'output': datasets.Value('string')
             }),
             # Homepage of the module for documentation
             homepage="http://module.homepage",
@@ -87,35 +87,30 @@ class charmatch(evaluate.Metric):
         # TODO: Download external resources if needed
         pass
 
-    def _compute(inputs, expected, outputs):
-        def get_score(input, expected, output):
-            print(input, expected, output)
-            deduped = {input, expected, output}
-            if len(deduped) == 1:
+    def _compute(input, expected, output):
+        print(input, expected, output)
+        deduped = {input, expected, output}
+        if len(deduped) == 1:
+            return 1.0
+        elif len(deduped) == 2:
+            if expected == output:
                 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}')
+                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}')
+            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
+            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 {
-            "fscore": avg
+            "fscore": f_05
         }