Nix commited on
Commit
60f81cf
1 Parent(s): d86fa34

Upload CLUTRR_v1.py

Browse files
Files changed (1) hide show
  1. CLUTRR_v1.py +153 -0
CLUTRR_v1.py ADDED
@@ -0,0 +1,153 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # -*- coding: utf-8 -*-
2
+ """CLUTRR_Dataset Loading Script.ipynb
3
+
4
+ Automatically generated by Colaboratory.
5
+
6
+ Original file is located at
7
+ https://colab.research.google.com/drive/1q9DdeHA5JbgTHkH6kfZe_KWHQOwHZA97
8
+ """
9
+
10
+ # coding=utf-8
11
+ # Copyright 2019 The CLUTRR Datasets Authors and the HuggingFace Datasets Authors.
12
+ #
13
+ # CLUTRR is CC-BY-NC 4.0 (Attr Non-Commercial Inter.) licensed, as found in the LICENSE file.
14
+ #
15
+ # Unless required by applicable law or agreed to in writing, software
16
+ # distributed under the License is distributed on an "AS IS" BASIS,
17
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
18
+ # See the License for the specific language governing permissions and
19
+ # limitations under the License.
20
+
21
+ # Lint as: python3
22
+ """The CLUTRR (Compositional Language Understanding and Text-based Relational Reasoning) benchmark."""
23
+
24
+
25
+ import csv
26
+ import os
27
+ import textwrap
28
+
29
+ import numpy as np
30
+
31
+ import datasets
32
+
33
+
34
+ _CLUTRR_CITATION = """\
35
+ @article{sinha2019clutrr,
36
+ Author = {Koustuv Sinha and Shagun Sodhani and Jin Dong and Joelle Pineau and William L. Hamilton},
37
+ Title = {CLUTRR: A Diagnostic Benchmark for Inductive Reasoning from Text},
38
+ Year = {2019},
39
+ journal = {Empirical Methods of Natural Language Processing (EMNLP)},
40
+ arxiv = {1908.06177}
41
+ }
42
+ """
43
+
44
+ _CLUTRR_DESCRIPTION = """\
45
+ CLUTRR (Compositional Language Understanding and Text-based Relational Reasoning),
46
+ a diagnostic benchmark suite, is first introduced in (https://arxiv.org/abs/1908.06177)
47
+ to test the systematic generalization and inductive reasoning capabilities of NLU systems.
48
+
49
+ """
50
+ _URL = "https://github.com/kliang5/CLUTRR_huggingface_dataset/tree/main/"
51
+ _TASK = ["gen_train23_test2to10", "gen_train234_test2to10", "rob_train_clean_23_test_all_23", "rob_train_disc_23_test_all_23", "rob_train_irr_23_test_all_23","rob_train_sup_23_test_all_23"]
52
+
53
+ class CLUTRR_v1(datasets.GeneratorBasedBuilder):
54
+ """BuilderConfig for CLUTRR."""
55
+
56
+ BUILDER_CONFIGS = [
57
+ datasets.BuilderConfig(
58
+ name=task,
59
+ version=datasets.Version("1.0.0"),
60
+ description="",
61
+ )
62
+ for task in _TASK
63
+ ]
64
+
65
+ def _info(self):
66
+ return datasets.DatasetInfo(
67
+ description=_CLUTRR_DESCRIPTION,
68
+ features=datasets.Features(
69
+ {
70
+ "id": datasets.Value("string"),
71
+ "story": datasets.Value("string"),
72
+ "query": datasets.Value("string"),
73
+ "target": datasets.Value("string"),
74
+ "clean_story": datasets.Value("string"),
75
+ "proof_state": datasets.Value("string"),
76
+ "f_comb": datasets.Value("string"),
77
+ "task_name": datasets.Value("string"),
78
+ "story_edges": datasets.Value("string"),
79
+ "edge_types": datasets.Value("string"),
80
+ "query_edge": datasets.Value("string"),
81
+ "genders": datasets.Value("string"),
82
+ "task_split": datasets.Value("string"),
83
+ }
84
+ ),
85
+ # No default supervised_keys (as we have to pass both premise
86
+ # and hypothesis as input).
87
+ supervised_keys=None,
88
+ homepage="https://www.cs.mcgill.ca/~ksinha4/clutrr/",
89
+ citation=_CLUTRR_CITATION,
90
+ )
91
+
92
+ def _split_generators(self, dl_manager):
93
+ """Returns SplitGenerators."""
94
+ # dl_manager is a datasets.download.DownloadManager that can be used to
95
+ # download and extract URLs
96
+
97
+ task = str(self.config.name)
98
+ urls_to_download = {
99
+ "test": _URL + task + "/test.csv",
100
+ "train": _URL + task + "/train.csv",
101
+ "validation": _URL + task + "/validation.csv",
102
+ }
103
+ downloaded_files = dl_manager.download_and_extract(urls_to_download)
104
+
105
+ return [
106
+ datasets.SplitGenerator(
107
+ name=datasets.Split.TRAIN,
108
+ # These kwargs will be passed to _generate_examples
109
+ gen_kwargs={
110
+ "filepath": os.path.join(downloaded_files["train"], task + "_train.jsonl"),
111
+ "task": task,
112
+ },
113
+ ),
114
+ datasets.SplitGenerator(
115
+ name=datasets.Split.VALIDATION,
116
+ # These kwargs will be passed to _generate_examples
117
+ gen_kwargs={
118
+ "filepath": os.path.join(downloaded_files["validation"], task + "_val.jsonl"),
119
+ "task": task,
120
+ },
121
+ ),
122
+ datasets.SplitGenerator(
123
+ name=datasets.Split.TEST,
124
+ # These kwargs will be passed to _generate_examples
125
+ gen_kwargs={
126
+ "filepath": os.path.join(downloaded_files["test"], task + "_test.jsonl"),
127
+ "task": task,
128
+ },
129
+ ),
130
+ ]
131
+
132
+ def _generate_examples(self, filepath, task):
133
+ """Yields examples."""
134
+ with open(filepath, encoding="utf-8") as f:
135
+ i = 0
136
+ for line in f:
137
+ data = json.loads(line)
138
+ i += 1
139
+ yield i, {
140
+ "id": data["id"],
141
+ "story": data["story"],
142
+ "query": data["query"],
143
+ "target": data["target"],
144
+ "clean_story": data["clean_story"],
145
+ "proof_state": data["proof_state"],
146
+ "f_comb": data["f_comb"],
147
+ "task_name": data["task_name"],
148
+ "story_edges": data["story_edges"],
149
+ "edge_types": data["edge_types"],
150
+ "query_edge": data["query_edge"],
151
+ "genders": data["genders"],
152
+ "task_split": data["task_split"],
153
+ }