Upload metric.py with huggingface_hub
Browse files
metric.py
ADDED
|
@@ -0,0 +1,147 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#####
|
| 2 |
+
# imports for hf system:
|
| 3 |
+
#####
|
| 4 |
+
from .artifact import __file__ as _
|
| 5 |
+
from .blocks import __file__ as _
|
| 6 |
+
from .card import __file__ as _
|
| 7 |
+
from .catalog import __file__ as _
|
| 8 |
+
from .collections import __file__ as _
|
| 9 |
+
from .common import __file__ as _
|
| 10 |
+
from .file_utils import __file__ as _
|
| 11 |
+
|
| 12 |
+
# from .fusion import __file__
|
| 13 |
+
from .generator_utils import __file__ as _
|
| 14 |
+
from .instructions import __file__ as _
|
| 15 |
+
from .loaders import __file__ as _
|
| 16 |
+
from .load import __file__ as _
|
| 17 |
+
from .metrics import __file__ as _
|
| 18 |
+
from .normalizers import __file__ as _
|
| 19 |
+
from .operator import __file__ as _
|
| 20 |
+
from .operators import __file__ as _
|
| 21 |
+
from .processors import __file__ as _
|
| 22 |
+
from .recipe import __file__ as _
|
| 23 |
+
from .register import __file__ as _
|
| 24 |
+
from .splitters import __file__ as _
|
| 25 |
+
from .split_utils import __file__ as _
|
| 26 |
+
from .stream import __file__ as _
|
| 27 |
+
from .task import __file__ as _
|
| 28 |
+
from .templates import __file__ as _
|
| 29 |
+
from .text_utils import __file__ as _
|
| 30 |
+
from .schema import __file__ as _
|
| 31 |
+
|
| 32 |
+
# from .utilize import __file__ as _
|
| 33 |
+
# from .validate import __file__ as _
|
| 34 |
+
#############
|
| 35 |
+
|
| 36 |
+
from .stream import MultiStream, Stream
|
| 37 |
+
|
| 38 |
+
from .operator import SequntialOperator, SequntialOperatorInitilizer, MultiStreamOperator, StreamInitializerOperator
|
| 39 |
+
|
| 40 |
+
from .operators import (
|
| 41 |
+
ApplyValueOperatorsField,
|
| 42 |
+
ApplyStreamOperatorsField,
|
| 43 |
+
SplitByValue,
|
| 44 |
+
MergeStreams,
|
| 45 |
+
FlattenInstances,
|
| 46 |
+
)
|
| 47 |
+
|
| 48 |
+
import evaluate
|
| 49 |
+
import datasets
|
| 50 |
+
|
| 51 |
+
from datasets import (
|
| 52 |
+
Features,
|
| 53 |
+
Value,
|
| 54 |
+
Sequence,
|
| 55 |
+
)
|
| 56 |
+
|
| 57 |
+
from dataclasses import field
|
| 58 |
+
from typing import List, Union, Dict, Optional, Generator, Any, Iterable
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
class MultiStreamScoreMean(MultiStreamOperator):
|
| 62 |
+
def aggegate_results(self, multi_stream: MultiStream):
|
| 63 |
+
scores = []
|
| 64 |
+
for stream in multi_stream.values():
|
| 65 |
+
instance = stream.peak()
|
| 66 |
+
scores.append(instance["score"]["global"]["score"])
|
| 67 |
+
|
| 68 |
+
from statistics import mean
|
| 69 |
+
|
| 70 |
+
return mean(scores)
|
| 71 |
+
|
| 72 |
+
def spread_results(self, stream: Stream, score: float):
|
| 73 |
+
for instance in stream:
|
| 74 |
+
instance["score"]["global"]["groups_mean_score"] = score
|
| 75 |
+
yield instance
|
| 76 |
+
|
| 77 |
+
def process(self, multi_stream: MultiStream) -> MultiStream:
|
| 78 |
+
mean_score = self.aggegate_results(multi_stream)
|
| 79 |
+
|
| 80 |
+
result = {}
|
| 81 |
+
for stream_name, stream in multi_stream.items():
|
| 82 |
+
result[stream_name] = Stream(self.spread_results, gen_kwargs={"stream": stream, "score": mean_score})
|
| 83 |
+
|
| 84 |
+
return MultiStream(result)
|
| 85 |
+
|
| 86 |
+
|
| 87 |
+
class FromPredictionsAndOriginalData(StreamInitializerOperator):
|
| 88 |
+
def zip(self, predictions, references):
|
| 89 |
+
for prediction, original in zip(predictions, references):
|
| 90 |
+
yield {**original, "prediction": prediction}
|
| 91 |
+
|
| 92 |
+
def process(self, predictions: List[str], references: Iterable, split_name: str = "all") -> MultiStream:
|
| 93 |
+
return MultiStream(
|
| 94 |
+
{split_name: Stream(self.zip, gen_kwargs={"predictions": predictions, "references": references})}
|
| 95 |
+
)
|
| 96 |
+
|
| 97 |
+
|
| 98 |
+
from .schema import UNITXT_DATASET_SCHEMA
|
| 99 |
+
|
| 100 |
+
|
| 101 |
+
class MetricRecipe(SequntialOperatorInitilizer):
|
| 102 |
+
def prepare(self):
|
| 103 |
+
self.steps = [
|
| 104 |
+
FromPredictionsAndOriginalData(),
|
| 105 |
+
ApplyValueOperatorsField(
|
| 106 |
+
value_field="prediction", operators_field="processors", default_operators=["to_string"]
|
| 107 |
+
),
|
| 108 |
+
SplitByValue(["group"]),
|
| 109 |
+
ApplyStreamOperatorsField(
|
| 110 |
+
"metrics",
|
| 111 |
+
reversed=True,
|
| 112 |
+
),
|
| 113 |
+
MultiStreamScoreMean(),
|
| 114 |
+
MergeStreams(),
|
| 115 |
+
]
|
| 116 |
+
|
| 117 |
+
|
| 118 |
+
UNITXT_METRIC_SCHEMA = Features({"predictions": Value("string"), "references": dict(UNITXT_DATASET_SCHEMA)})
|
| 119 |
+
|
| 120 |
+
|
| 121 |
+
# @evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
|
| 122 |
+
class UnitextMetric(evaluate.Metric):
|
| 123 |
+
def _info(self):
|
| 124 |
+
return evaluate.MetricInfo(
|
| 125 |
+
description="_DESCRIPTION",
|
| 126 |
+
citation="_CITATION",
|
| 127 |
+
# inputs_description=_KWARGS_DESCRIPTION,
|
| 128 |
+
features=UNITXT_METRIC_SCHEMA,
|
| 129 |
+
codebase_urls=["https://"],
|
| 130 |
+
reference_urls=[
|
| 131 |
+
"https://",
|
| 132 |
+
"https://",
|
| 133 |
+
],
|
| 134 |
+
)
|
| 135 |
+
|
| 136 |
+
def _compute(self, predictions: List[str], references: Iterable, flatten: bool = False, split_name: str = "all"):
|
| 137 |
+
recipe = MetricRecipe()
|
| 138 |
+
|
| 139 |
+
multi_stream = recipe(predictions=predictions, references=references, split_name=split_name)
|
| 140 |
+
|
| 141 |
+
if flatten:
|
| 142 |
+
operator = FlattenInstances()
|
| 143 |
+
multi_stream = operator(multi_stream)
|
| 144 |
+
|
| 145 |
+
stream = multi_stream[split_name]
|
| 146 |
+
|
| 147 |
+
return list(stream)
|