ClaraBing commited on
Commit
8453f96
1 Parent(s): 59926fb

add ace (MockupSampler)

Browse files
Files changed (1) hide show
  1. ace.py +164 -0
ace.py ADDED
@@ -0,0 +1,164 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2020 The HuggingFace Datasets Authors.
2
+ # Copyright 2023 Cyril Zhang.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+
17
+ import csv
18
+ import json
19
+ import os
20
+
21
+ import datasets
22
+ import numpy as np
23
+
24
+
25
+ _CITATION = """\
26
+ """
27
+
28
+ _DESCRIPTION = """\
29
+ Online dataset mockup.
30
+ """
31
+
32
+ _HOMEPAGE = ""
33
+
34
+ _LICENSE = ""
35
+
36
+ _URLS = {}
37
+
38
+ class MockupDataset(datasets.GeneratorBasedBuilder):
39
+ """TODO: Short description of my dataset."""
40
+
41
+ VERSION = datasets.Version("0.0.0")
42
+ BUILDER_CONFIGS = []
43
+
44
+ def __init__(self, name=None, data_config={}, **kwargs):
45
+ super().__init__(**kwargs)
46
+
47
+ """
48
+ Set default configs
49
+ """
50
+ if name is None:
51
+ name = 'parity'
52
+ if 'length' not in data_config:
53
+ data_config['length'] = 20
54
+ if 'size' not in data_config:
55
+ data_config['size'] = 100
56
+
57
+ self.data_config = data_config
58
+ # self.sampler = AutomatonSampler(name, data_config)
59
+ self.sampler = dataset_map[name](data_config)
60
+
61
+ def _info(self):
62
+ features = datasets.Features(
63
+ {
64
+ "x": datasets.Value("null"),
65
+ "y": datasets.Value("null")
66
+ }
67
+ )
68
+
69
+ return datasets.DatasetInfo(
70
+ description=_DESCRIPTION,
71
+ features=features,
72
+ homepage=_HOMEPAGE,
73
+ license=_LICENSE,
74
+ citation=_CITATION,
75
+ )
76
+
77
+ def _split_generators(self, dl_manager):
78
+ return [
79
+ datasets.SplitGenerator(
80
+ name=datasets.Split.TRAIN,
81
+ gen_kwargs={
82
+ "split": "train",
83
+ },
84
+ )
85
+ ]
86
+
87
+ # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
88
+ def _generate_examples(self, split):
89
+ for i in range(self.data_config['size']):
90
+ x, y = self.sampler.sample()
91
+ yield i, {
92
+ "x": x,
93
+ "y": y
94
+ }
95
+
96
+
97
+ class AutomatonSampler:
98
+ def __init__(self, data_config):
99
+ # self.name = name
100
+ self.data_config = data_config
101
+
102
+ if 'seed' in self.data_config:
103
+ self.np_rng = np.random.default_rng(self.data_config['seed'])
104
+ else:
105
+ self.np_rng = np.random.default_rng()
106
+
107
+ self.n_states = data_config['n_states']
108
+ self.T = self.data_config['length']
109
+
110
+ def f(self, x):
111
+ """
112
+ Get output sequence given an input seq
113
+ """
114
+ raise NotImplementedError()
115
+
116
+ def sample(self):
117
+ raise NotImplementedError()
118
+
119
+
120
+ class ParitySampler(AutomatonSampler):
121
+ def __init__(self, data_config):
122
+ super(ParitySampler, self).__init__(data_config)
123
+ self.name = 'parity'
124
+ self.data_config = data_config
125
+
126
+ def f(self, x):
127
+ return np.cumsum(x) % 2
128
+
129
+ def sample(self):
130
+ x = self.np_rng.binomial(1,0.5,size=self.T)
131
+ return x, self.f(x)
132
+
133
+
134
+ class FlipflopSampler(AutomatonSampler):
135
+ def __init__(self, data_config):
136
+ super(FlipflopSampler, self).__init__(data_config)
137
+ self.name = 'parity'
138
+ self.data_config = data_config
139
+
140
+ self.n_actions = self.n_states + 1
141
+ self.transition = np.array([list(range(self.n_actions))] + [[i+1]*self.n_actions for i in range(self.n_states)]).T
142
+
143
+ def f(self, x):
144
+ state, states = 0, []
145
+ for action in x:
146
+ state = self.transition[state, action]
147
+ states += state,
148
+ return np.array(states)
149
+
150
+ def sample(self):
151
+ rand = np.random.uniform(size=self.T)
152
+ nonzero_pos = (rand < 0.5).astype(np.int64)
153
+ writes = np.random.choice(range(1, self.n_states+1), size=self.T)
154
+ x = writes * nonzero_pos
155
+ return x, self.f(x)
156
+
157
+
158
+ dataset_map = {
159
+ 'parity': ParitySampler,
160
+ 'flipflop': FlipflopSampler,
161
+ # TODO: more datasets
162
+ }
163
+
164
+