kedudzic commited on
Commit
c68ac0f
1 Parent(s): 437600c

update compute

Browse files
Files changed (1) hide show
  1. charmatch.py +25 -20
charmatch.py CHANGED
@@ -87,30 +87,35 @@ class charmatch(evaluate.Metric):
87
  # TODO: Download external resources if needed
88
  pass
89
 
90
- def _compute(self, input, expected, output):
91
- print(input, expected, output)
92
- deduped = {input, expected, output}
93
- if len(deduped) == 1:
94
- return 1.0
95
- elif len(deduped) == 2:
96
- if expected == output:
97
  return 1.0
 
 
 
 
 
98
  else:
99
- return 0.0
100
- else:
101
- expected_corrections = lev(input, expected)
102
- distance_to_input = lev(input, output)
103
- distance_to_expected = lev(output, expected)
104
- print(f'dl(s,g): {expected_corrections}\ndl(s,h): {distance_to_input}\ndl(h,g): {distance_to_expected}')
105
 
106
- true_positives = min(expected_corrections, max(0, (expected_corrections + distance_to_input - distance_to_expected))) / 2
107
- print(f'T: {true_positives}')
108
 
109
- precision = true_positives / distance_to_input
110
- recall = true_positives / expected_corrections
111
- f_05 = (1 + 0.5**2) * (precision * recall) / (0.5**2 * precision + recall)
112
- print(f'P: {precision}\nR: {recall}')
 
 
 
 
113
 
114
  return {
115
- "fscore": f_05
116
  }
 
87
  # TODO: Download external resources if needed
88
  pass
89
 
90
+ def _compute(self, inputs, expected, outputs):
91
+ def get_score(input, expected, output):
92
+ print(input, expected, output)
93
+ deduped = {input, expected, output}
94
+ if len(deduped) == 1:
 
 
95
  return 1.0
96
+ elif len(deduped) == 2:
97
+ if expected == output:
98
+ return 1.0
99
+ else:
100
+ return 0.0
101
  else:
102
+ expected_corrections = lev(input, expected)
103
+ distance_to_input = lev(input, output)
104
+ distance_to_expected = lev(output, expected)
105
+ print(f'dl(s,g): {expected_corrections}\ndl(s,h): {distance_to_input}\ndl(h,g): {distance_to_expected}')
 
 
106
 
107
+ true_positives = min(expected_corrections, max(0, (expected_corrections + distance_to_input - distance_to_expected))) / 2
108
+ print(f'T: {true_positives}')
109
 
110
+ precision = true_positives / distance_to_input
111
+ recall = true_positives / expected_corrections
112
+ f_05 = (1 + 0.5**2) * (precision * recall) / (0.5**2 * precision + recall)
113
+ print(f'P: {precision}\nR: {recall}')
114
+
115
+ return f_05
116
+
117
+ avg = sum([get_score(*row) for row in zip(inputs, expected, outputs)]) / len(inputs) * 100
118
 
119
  return {
120
+ "fscore": avg
121
  }