File size: 5,316 Bytes
b783bce
 
 
 
 
 
 
e5a087b
 
 
 
 
 
6f3c593
c4dd600
e5a087b
f32487c
b783bce
e5a087b
6798e06
e5a087b
 
b783bce
e5a087b
 
b783bce
 
 
 
 
 
e5a087b
b783bce
f61d613
b783bce
 
 
 
 
e5a087b
 
c4dd600
e5a087b
 
b783bce
f32487c
b783bce
e5a087b
b783bce
f32487c
b783bce
e5a087b
 
 
 
6798e06
b783bce
 
6452fbf
e5a087b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5818152
 
e5a087b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b783bce
e5a087b
 
f61d613
 
468c32e
f61d613
f32487c
e5a087b
 
 
 
 
 
 
 
 
 
 
 
 
 
6798e06
c4dd600
6798e06
c4dd600
6798e06
c4dd600
 
 
6798e06
c4dd600
 
 
6798e06
 
 
e5a087b
f95da7e
5818152
 
e5a087b
 
 
 
 
 
 
 
 
5818152
 
e5a087b
6798e06
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
from dataclasses import field
from typing import Any, Dict, Generator, Iterable, List, Optional, Union

import datasets
import evaluate
from datasets import Features, Sequence, Value

from .artifact import __file__ as _
from .blocks import __file__ as _
from .card import __file__ as _
from .catalog import __file__ as _
from .collections import __file__ as _
from .common import __file__ as _
from .dataclass import __file__ as _
from .dict_utils import __file__ as _
from .file_utils import __file__ as _
from .formats import __file__ as _
from .fusion import __file__ as _
from .generator_utils import __file__ as _
from .hf_utils import __file__ as _
from .instructions import __file__ as _
from .load import __file__ as _
from .loaders import __file__ as _
from .metrics import __file__ as _
from .normalizers import __file__ as _
from .operator import (
    MultiStreamOperator,
    SequntialOperator,
    SequntialOperatorInitilizer,
    StreamInitializerOperator,
)
from .operator import __file__ as _
from .operators import (
    ApplyOperatorsField,
    ApplyStreamOperatorsField,
    FlattenInstances,
    MergeStreams,
    SplitByValue,
)
from .operators import __file__ as _
from .processors import __file__ as _
from .random_utils import __file__ as _
from .recipe import __file__ as _
from .register import __file__ as _
from .register import register_all_artifacts
from .renderers import __file__ as _
from .schema import __file__ as _
from .split_utils import __file__ as _
from .splitters import __file__ as _
from .standard import __file__ as _
from .stream import MultiStream, Stream
from .stream import __file__ as _
from .task import __file__ as _
from .templates import __file__ as _
from .text_utils import __file__ as _
from .type_utils import __file__ as _
from .utils import __file__ as _
from .validate import __file__ as _
from .version import __file__ as _


class MultiStreamScoreMean(MultiStreamOperator):
    def aggegate_results(self, multi_stream: MultiStream):
        scores = []
        for stream in multi_stream.values():
            instance = stream.peak()
            scores.append(instance["score"]["global"]["score"])

        from statistics import mean

        return mean(scores)

    def spread_results(self, stream: Stream, score: float):
        for instance in stream:
            instance["score"]["global"]["groups_mean_score"] = score
            yield instance

    def process(self, multi_stream: MultiStream) -> MultiStream:
        mean_score = self.aggegate_results(multi_stream)

        result = {}
        for stream_name, stream in multi_stream.items():
            result[stream_name] = Stream(self.spread_results, gen_kwargs={"stream": stream, "score": mean_score})

        return MultiStream(result)


class FromPredictionsAndOriginalData(StreamInitializerOperator):
    def zip(self, predictions, references):
        for prediction, original in zip(predictions, references):
            yield {**original, "prediction": prediction}

    def process(self, predictions: List[str], references: Iterable, split_name: str = "all") -> MultiStream:
        return MultiStream(
            {split_name: Stream(self.zip, gen_kwargs={"predictions": predictions, "references": references})}
        )


from .schema import UNITXT_DATASET_SCHEMA


class MetricRecipe(SequntialOperatorInitilizer):
    def prepare(self):
        register_all_artifacts()
        self.steps = [
            FromPredictionsAndOriginalData(),
            ApplyOperatorsField(
                inputs_fields=["prediction", "references"],
                fields_to_treat_as_list=["references"],
                operators_field="postprocessors",
                default_operators=["processors.to_string_stripped"],
            ),
            SplitByValue(["group"]),
            ApplyStreamOperatorsField(
                "metrics",
                reversed=True,
            ),
            MultiStreamScoreMean(),
            MergeStreams(),
        ]


UNITXT_METRIC_SCHEMA = Features({"predictions": Value("string"), "references": dict(UNITXT_DATASET_SCHEMA)})


def _compute(predictions: List[str], references: Iterable, flatten: bool = False, split_name: str = "all"):
    recipe = MetricRecipe()

    multi_stream = recipe(predictions=predictions, references=references, split_name=split_name)

    if flatten:
        operator = FlattenInstances()
        multi_stream = operator(multi_stream)

    stream = multi_stream[split_name]

    return list(stream)


# TODO: currently we have two classes with this name. metric.Metric and matrics.Metric...
# @evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
class Metric(evaluate.Metric):
    def _info(self):
        return evaluate.MetricInfo(
            description="_DESCRIPTION",
            citation="_CITATION",
            # inputs_description=_KWARGS_DESCRIPTION,
            features=UNITXT_METRIC_SCHEMA,
            codebase_urls=["https://"],
            reference_urls=[
                "https://",
                "https://",
            ],
        )

    def _compute(self, predictions: List[str], references: Iterable, flatten: bool = False, split_name: str = "all"):
        return _compute(predictions=predictions, references=references, flatten=flatten, split_name=split_name)