Upload splitters.py with huggingface_hub
Browse files- splitters.py +57 -12
splitters.py
CHANGED
@@ -1,10 +1,11 @@
|
|
1 |
import itertools
|
2 |
from abc import abstractmethod
|
|
|
3 |
from typing import Dict, List
|
4 |
|
5 |
from .artifact import Artifact
|
6 |
from .operator import InstanceOperatorWithMultiStreamAccess, MultiStreamOperator
|
7 |
-
from .random_utils import
|
8 |
from .split_utils import (
|
9 |
parse_random_mix_string,
|
10 |
parse_slices_string,
|
@@ -82,6 +83,7 @@ class SliceSplit(Splitter):
|
|
82 |
|
83 |
class Sampler(Artifact):
|
84 |
sample_size: int = None
|
|
|
85 |
|
86 |
def prepare(self):
|
87 |
super().prepare()
|
@@ -95,6 +97,11 @@ class Sampler(Artifact):
|
|
95 |
size = int(size)
|
96 |
self.sample_size = size
|
97 |
|
|
|
|
|
|
|
|
|
|
|
98 |
@abstractmethod
|
99 |
def sample(
|
100 |
self, instances_pool: List[Dict[str, object]]
|
@@ -107,22 +114,52 @@ class RandomSampler(Sampler):
|
|
107 |
self, instances_pool: List[Dict[str, object]]
|
108 |
) -> List[Dict[str, object]]:
|
109 |
instances_pool = list(instances_pool)
|
110 |
-
return
|
111 |
|
112 |
|
113 |
class DiverseLabelsSampler(Sampler):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
114 |
choices: str = "choices"
|
|
|
115 |
|
116 |
def prepare(self):
|
117 |
super().prepare()
|
118 |
-
self.
|
119 |
|
120 |
def examplar_repr(self, examplar):
|
121 |
if "inputs" not in examplar:
|
122 |
raise ValueError(f"'inputs' field is missing from '{examplar}'.")
|
123 |
inputs = examplar["inputs"]
|
124 |
if self.choices not in inputs:
|
125 |
-
raise ValueError(f"{self.choices} field is missing from '{inputs}'.")
|
126 |
choices = inputs[self.choices]
|
127 |
if not isinstance(choices, list):
|
128 |
raise ValueError(
|
@@ -131,7 +168,11 @@ class DiverseLabelsSampler(Sampler):
|
|
131 |
|
132 |
if "outputs" not in examplar:
|
133 |
raise ValueError(f"'outputs' field is missing from '{examplar}'.")
|
134 |
-
|
|
|
|
|
|
|
|
|
135 |
if not isinstance(examplar_outputs, list):
|
136 |
raise ValueError(
|
137 |
f"Unexpected examplar_outputs value '{examplar_outputs}'. Expected a list."
|
@@ -151,19 +192,23 @@ class DiverseLabelsSampler(Sampler):
|
|
151 |
def sample(
|
152 |
self, instances_pool: List[Dict[str, object]]
|
153 |
) -> List[Dict[str, object]]:
|
154 |
-
if self.
|
155 |
-
self.
|
156 |
-
all_labels = list(self.
|
157 |
-
|
158 |
from collections import Counter
|
159 |
|
|
|
|
|
|
|
|
|
160 |
total_allocated = 0
|
161 |
allocations = Counter()
|
162 |
|
163 |
while total_allocated < self.sample_size:
|
164 |
for label in all_labels:
|
165 |
if total_allocated < self.sample_size:
|
166 |
-
if len(self.
|
167 |
allocations[label] += 1
|
168 |
total_allocated += 1
|
169 |
else:
|
@@ -171,10 +216,10 @@ class DiverseLabelsSampler(Sampler):
|
|
171 |
|
172 |
result = []
|
173 |
for label, allocation in allocations.items():
|
174 |
-
sample =
|
175 |
result.extend(sample)
|
176 |
|
177 |
-
|
178 |
return result
|
179 |
|
180 |
|
|
|
1 |
import itertools
|
2 |
from abc import abstractmethod
|
3 |
+
from random import Random
|
4 |
from typing import Dict, List
|
5 |
|
6 |
from .artifact import Artifact
|
7 |
from .operator import InstanceOperatorWithMultiStreamAccess, MultiStreamOperator
|
8 |
+
from .random_utils import new_random_generator
|
9 |
from .split_utils import (
|
10 |
parse_random_mix_string,
|
11 |
parse_slices_string,
|
|
|
83 |
|
84 |
class Sampler(Artifact):
|
85 |
sample_size: int = None
|
86 |
+
random_generator: Random = new_random_generator(sub_seed="Sampler")
|
87 |
|
88 |
def prepare(self):
|
89 |
super().prepare()
|
|
|
97 |
size = int(size)
|
98 |
self.sample_size = size
|
99 |
|
100 |
+
def init_new_random_generator(self):
|
101 |
+
self.random_generator = new_random_generator(
|
102 |
+
sub_seed="init_new_random_generator"
|
103 |
+
)
|
104 |
+
|
105 |
@abstractmethod
|
106 |
def sample(
|
107 |
self, instances_pool: List[Dict[str, object]]
|
|
|
114 |
self, instances_pool: List[Dict[str, object]]
|
115 |
) -> List[Dict[str, object]]:
|
116 |
instances_pool = list(instances_pool)
|
117 |
+
return self.random_generator.sample(instances_pool, self.sample_size)
|
118 |
|
119 |
|
120 |
class DiverseLabelsSampler(Sampler):
|
121 |
+
"""Selects a balanced sample of instances based on an output field.
|
122 |
+
|
123 |
+
(used for selecting demonstrations in-context learning)
|
124 |
+
|
125 |
+
The field must contain list of values e.g ['dog'], ['cat'], ['dog','cat','cow'].
|
126 |
+
The balancing is done such that each value or combination of values
|
127 |
+
appears as equals as possible in the samples.
|
128 |
+
|
129 |
+
The `choices` param is required and determines which values should be considered.
|
130 |
+
|
131 |
+
Example:
|
132 |
+
If choices is ['dog,'cat'] , then the following combinations will be considered.
|
133 |
+
['']
|
134 |
+
['cat']
|
135 |
+
['dog']
|
136 |
+
['dog','cat']
|
137 |
+
|
138 |
+
If the instance contains a value not in the 'choice' param, it is ignored. For example,
|
139 |
+
if choices is ['dog,'cat'] and the instance field is ['dog','cat','cow'], then 'cow' is ignored
|
140 |
+
then the instance is considered as ['dog','cat'].
|
141 |
+
|
142 |
+
Args:
|
143 |
+
sample_size - number of samples to extract
|
144 |
+
choices - name of input field that contains the list of values to balance on
|
145 |
+
labels - name of output field with labels that must be balanced
|
146 |
+
|
147 |
+
|
148 |
+
"""
|
149 |
+
|
150 |
choices: str = "choices"
|
151 |
+
labels: str = "labels"
|
152 |
|
153 |
def prepare(self):
|
154 |
super().prepare()
|
155 |
+
self.labels_cache = None
|
156 |
|
157 |
def examplar_repr(self, examplar):
|
158 |
if "inputs" not in examplar:
|
159 |
raise ValueError(f"'inputs' field is missing from '{examplar}'.")
|
160 |
inputs = examplar["inputs"]
|
161 |
if self.choices not in inputs:
|
162 |
+
raise ValueError(f"'{self.choices}' field is missing from '{inputs}'.")
|
163 |
choices = inputs[self.choices]
|
164 |
if not isinstance(choices, list):
|
165 |
raise ValueError(
|
|
|
168 |
|
169 |
if "outputs" not in examplar:
|
170 |
raise ValueError(f"'outputs' field is missing from '{examplar}'.")
|
171 |
+
outputs = examplar["outputs"]
|
172 |
+
if self.labels not in outputs:
|
173 |
+
raise ValueError(f"'{self.labels}' field is missing from '{outputs}'.")
|
174 |
+
|
175 |
+
examplar_outputs = examplar["outputs"][self.labels]
|
176 |
if not isinstance(examplar_outputs, list):
|
177 |
raise ValueError(
|
178 |
f"Unexpected examplar_outputs value '{examplar_outputs}'. Expected a list."
|
|
|
192 |
def sample(
|
193 |
self, instances_pool: List[Dict[str, object]]
|
194 |
) -> List[Dict[str, object]]:
|
195 |
+
if self.labels_cache is None:
|
196 |
+
self.labels_cache = self.divide_by_repr(instances_pool)
|
197 |
+
all_labels = list(self.labels_cache.keys())
|
198 |
+
self.random_generator.shuffle(all_labels)
|
199 |
from collections import Counter
|
200 |
|
201 |
+
if self.sample_size > len(instances_pool):
|
202 |
+
raise ValueError(
|
203 |
+
f"Request sample size {self.sample_size} is greater than number of instances {len(instances_pool)}"
|
204 |
+
)
|
205 |
total_allocated = 0
|
206 |
allocations = Counter()
|
207 |
|
208 |
while total_allocated < self.sample_size:
|
209 |
for label in all_labels:
|
210 |
if total_allocated < self.sample_size:
|
211 |
+
if len(self.labels_cache[label]) - allocations[label] > 0:
|
212 |
allocations[label] += 1
|
213 |
total_allocated += 1
|
214 |
else:
|
|
|
216 |
|
217 |
result = []
|
218 |
for label, allocation in allocations.items():
|
219 |
+
sample = self.random_generator.sample(self.labels_cache[label], allocation)
|
220 |
result.extend(sample)
|
221 |
|
222 |
+
self.random_generator.shuffle(result)
|
223 |
return result
|
224 |
|
225 |
|