Datasets:

ArXiv:
Elron commited on
Commit
7d7cc9e
1 Parent(s): 455ee62

Upload metrics.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. metrics.py +141 -0
metrics.py ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from .stream import Stream
2
+ from .operator import SingleStreamOperator, StreamInstanceOperator
3
+ from dataclasses import dataclass, field
4
+ from abc import abstractmethod, ABC
5
+
6
+ from typing import List, Dict, Any
7
+
8
+
9
+ def absrtact_factory():
10
+ return {}
11
+
12
+
13
+ def abstract_field():
14
+ return field(default_factory=absrtact_factory)
15
+
16
+
17
+ class UpdateStream(StreamInstanceOperator):
18
+ update: dict
19
+
20
+ def process(self, instance: Dict[str, Any], stream_name: str = None) -> Dict[str, Any]:
21
+ instance.update(self.update)
22
+ return instance
23
+
24
+
25
+ class Metric(ABC):
26
+ @property
27
+ @abstractmethod
28
+ def main_score(self):
29
+ pass
30
+
31
+
32
+ class GlobalMetric(SingleStreamOperator, Metric):
33
+ def process(self, stream: Stream):
34
+ references = []
35
+ predictions = []
36
+ global_score = {}
37
+
38
+ instances = []
39
+
40
+ for instance in stream:
41
+ if "score" not in instance:
42
+ instance["score"] = {"global": global_score, "instance": {}}
43
+ else:
44
+ global_score = instance["score"]["global"]
45
+
46
+ refs, pred = instance["references"], instance["prediction"]
47
+
48
+ instance_score = self._compute([refs], [pred])
49
+ instance["score"]["instance"].update(instance_score)
50
+
51
+ references.append(refs)
52
+ predictions.append(pred)
53
+ instances.append(instance)
54
+
55
+ result = self._compute(references, predictions)
56
+
57
+ global_score.update(result)
58
+
59
+ for instance in instances:
60
+ instance["score"]["global"] = global_score
61
+ yield instance
62
+
63
+ def _compute(self, references: List[List[str]], predictions: List[str]) -> dict:
64
+ result = self.compute(references, predictions)
65
+ result["score"] = result[self.main_score]
66
+ return result
67
+
68
+ @abstractmethod
69
+ def compute(self, references: List[List[str]], predictions: List[str]) -> dict:
70
+ pass
71
+
72
+
73
+ class InstanceMetric(SingleStreamOperator, Metric):
74
+ implemented_reductions: List[str] = field(default_factory=lambda: ["mean"])
75
+
76
+ @property
77
+ @abstractmethod
78
+ def reduction_map(self) -> dict:
79
+ pass
80
+
81
+ def process(self, stream: Stream):
82
+ global_score = {}
83
+ instances = []
84
+
85
+ for instance in stream:
86
+ refs, pred = instance["references"], instance["prediction"]
87
+
88
+ instance_score = self._compute(refs, pred)
89
+
90
+ if "score" not in instance:
91
+ instance["score"] = {"global": global_score, "instance": {}}
92
+ else:
93
+ global_score = instance["score"]["global"]
94
+
95
+ instance["score"]["instance"].update(instance_score)
96
+
97
+ instances.append(instance)
98
+
99
+ for reduction, fields in self.reduction_map.items():
100
+ assert (
101
+ reduction in self.implemented_reductions
102
+ ), f"Reduction {reduction} is not implemented, use one of {self.implemented_reductions}"
103
+
104
+ if reduction == "mean":
105
+ from statistics import mean
106
+
107
+ for field in fields:
108
+ global_score[field] = mean([instance["score"]["instance"][field] for instance in instances])
109
+ if field == self.main_score:
110
+ global_score["score"] = global_score[field]
111
+
112
+ for instance in instances:
113
+ yield instance
114
+
115
+ def _compute(self, references: List[List[str]], predictions: List[str]) -> dict:
116
+ result = self.compute(references, predictions)
117
+ result["score"] = result[self.main_score]
118
+ return result
119
+
120
+ @abstractmethod
121
+ def compute(self, references: List[str], prediction: str) -> dict:
122
+ pass
123
+
124
+
125
+ class SingleReferenceInstanceMetric(InstanceMetric):
126
+ def _compute(self, references: List[str], prediction: str) -> dict:
127
+ result = self.compute(references[0], prediction)
128
+ result["score"] = result[self.main_score]
129
+ return result
130
+
131
+ @abstractmethod
132
+ def compute(self, reference, prediction: str) -> dict:
133
+ pass
134
+
135
+
136
+ class Accuracy(SingleReferenceInstanceMetric):
137
+ reduction_map = {"mean": ["accuracy"]}
138
+ main_score = "accuracy"
139
+
140
+ def compute(self, reference, prediction: str) -> dict:
141
+ return {"accuracy": float(str(reference) == str(prediction))}