Nix commited on
Commit
28bbe40
1 Parent(s): d023b4d

Create CLUTRR_v1.0.py

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