Upload splitters.py with huggingface_hub
Browse files- splitters.py +123 -0
splitters.py
ADDED
|
@@ -0,0 +1,123 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from .stream import MultiStream
|
| 2 |
+
from .operator import MultiStreamOperator, InstanceOperatorWithGlobalAccess
|
| 3 |
+
from .generator_utils import ReusableGenerator
|
| 4 |
+
from .artifact import Artifact
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
from typing import Optional, Dict, List
|
| 8 |
+
from dataclasses import field
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
class Splitter(MultiStreamOperator):
|
| 12 |
+
pass
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
import random
|
| 16 |
+
|
| 17 |
+
from .split_utils import (
|
| 18 |
+
parse_random_mix_string,
|
| 19 |
+
random_mix_streams,
|
| 20 |
+
parse_slices_string,
|
| 21 |
+
slice_streams,
|
| 22 |
+
)
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
class SplitRandomMix(Splitter):
|
| 26 |
+
mix: Dict[str, str]
|
| 27 |
+
|
| 28 |
+
def process(self, multi_stream: MultiStream) -> MultiStream:
|
| 29 |
+
mapping = {k: parse_random_mix_string(v) for k, v in self.mix.items()}
|
| 30 |
+
generators = random_mix_streams(multi_stream, mapping)
|
| 31 |
+
return MultiStream.from_generators(generators, streaming=True)
|
| 32 |
+
|
| 33 |
+
|
| 34 |
+
class SliceSplit(Splitter):
|
| 35 |
+
slices: Dict[str, str]
|
| 36 |
+
|
| 37 |
+
def process(self, multi_stream: MultiStream) -> MultiStream:
|
| 38 |
+
mapping = {k: parse_slices_string(v) for k, v in self.slices.items()}
|
| 39 |
+
generators = slice_streams(multi_stream, mapping)
|
| 40 |
+
return MultiStream.from_generators(generators, streaming=True)
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
class Sampler(Artifact):
|
| 44 |
+
sample_size: int
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
class RandomSampler(Sampler):
|
| 48 |
+
def sample(self, instances_pool: List[Dict[str, object]]) -> List[Dict[str, object]]:
|
| 49 |
+
instances_pool = list(instances_pool)
|
| 50 |
+
return random.sample(instances_pool, self.sample_size)
|
| 51 |
+
|
| 52 |
+
|
| 53 |
+
class SpreadSplit(InstanceOperatorWithGlobalAccess):
|
| 54 |
+
source_stream: str = None
|
| 55 |
+
target_field: str = None
|
| 56 |
+
sampler: Sampler = None
|
| 57 |
+
|
| 58 |
+
def prepare(self):
|
| 59 |
+
self.accessible_streams = [self.source_stream]
|
| 60 |
+
self.cache_accessible_streams = True
|
| 61 |
+
self.local_cache = None
|
| 62 |
+
|
| 63 |
+
def verify(self):
|
| 64 |
+
assert self.source_stream is not None, "Source stream must be specified"
|
| 65 |
+
assert self.target_field is not None, "Target field must be specified"
|
| 66 |
+
assert self.sampler is not None, "Sampler must be specified"
|
| 67 |
+
return super().verify()
|
| 68 |
+
|
| 69 |
+
def process(self, instance: Dict[str, object], multi_stream: MultiStream) -> Dict[str, object]:
|
| 70 |
+
if self.local_cache is None:
|
| 71 |
+
self.local_cache = list(multi_stream[self.source_stream])
|
| 72 |
+
|
| 73 |
+
source_stream = self.local_cache
|
| 74 |
+
|
| 75 |
+
sampled_instances = self.sampler.sample(source_stream)
|
| 76 |
+
instance[self.target_field] = sampled_instances
|
| 77 |
+
return instance
|
| 78 |
+
|
| 79 |
+
|
| 80 |
+
if __name__ == "__main__":
|
| 81 |
+
# some tests
|
| 82 |
+
import random
|
| 83 |
+
|
| 84 |
+
random.seed(0)
|
| 85 |
+
splitter = SplitRandomMix(
|
| 86 |
+
mix={
|
| 87 |
+
"train": "train[90%]+validation[50%]",
|
| 88 |
+
"validation": "train[10%]+validation[50%]",
|
| 89 |
+
"test": "test",
|
| 90 |
+
}
|
| 91 |
+
)
|
| 92 |
+
|
| 93 |
+
def generator(name, size):
|
| 94 |
+
for i in range(size):
|
| 95 |
+
yield {"text": f"{name}_{i}"}
|
| 96 |
+
|
| 97 |
+
stream = MultiStream.from_generators(
|
| 98 |
+
{
|
| 99 |
+
"train": ReusableGenerator(generator, gen_kwargs={"name": "train", "size": 10}),
|
| 100 |
+
"validation": ReusableGenerator(generator, gen_kwargs={"name": "validation", "size": 10}),
|
| 101 |
+
"test": ReusableGenerator(generator, gen_kwargs={"name": "test", "size": 10}),
|
| 102 |
+
}
|
| 103 |
+
)
|
| 104 |
+
|
| 105 |
+
ds = splitter(stream)
|
| 106 |
+
for key, value in ds.items():
|
| 107 |
+
print(key)
|
| 108 |
+
for item in value:
|
| 109 |
+
print(item)
|
| 110 |
+
|
| 111 |
+
splitter = SliceSplit(
|
| 112 |
+
slices={
|
| 113 |
+
"train": "train[:2]+train[2:4]",
|
| 114 |
+
"validation": "train[4:6]",
|
| 115 |
+
"test": "train[6:]+test",
|
| 116 |
+
}
|
| 117 |
+
)
|
| 118 |
+
|
| 119 |
+
ds = splitter(stream)
|
| 120 |
+
for key, value in ds.items():
|
| 121 |
+
print(key)
|
| 122 |
+
for item in value:
|
| 123 |
+
print(item)
|