Datasets:
Tasks:
Text2Text Generation
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
10K - 100K
ArXiv:
Tags:
concepts-to-text
License:
Commit
•
4c592de
1
Parent(s):
6bcab1c
Delete loading script
Browse files- common_gen.py +0 -109
common_gen.py
DELETED
@@ -1,109 +0,0 @@
|
|
1 |
-
import json
|
2 |
-
import os
|
3 |
-
import random
|
4 |
-
|
5 |
-
import datasets
|
6 |
-
|
7 |
-
|
8 |
-
random.seed(42) # This is important, to ensure the same order for concept sets as the official script.
|
9 |
-
|
10 |
-
_CITATION = """\
|
11 |
-
@inproceedings{lin-etal-2020-commongen,
|
12 |
-
title = "{C}ommon{G}en: A Constrained Text Generation Challenge for Generative Commonsense Reasoning",
|
13 |
-
author = "Lin, Bill Yuchen and
|
14 |
-
Zhou, Wangchunshu and
|
15 |
-
Shen, Ming and
|
16 |
-
Zhou, Pei and
|
17 |
-
Bhagavatula, Chandra and
|
18 |
-
Choi, Yejin and
|
19 |
-
Ren, Xiang",
|
20 |
-
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2020",
|
21 |
-
month = nov,
|
22 |
-
year = "2020",
|
23 |
-
address = "Online",
|
24 |
-
publisher = "Association for Computational Linguistics",
|
25 |
-
url = "https://www.aclweb.org/anthology/2020.findings-emnlp.165",
|
26 |
-
doi = "10.18653/v1/2020.findings-emnlp.165",
|
27 |
-
pages = "1823--1840"
|
28 |
-
}
|
29 |
-
"""
|
30 |
-
|
31 |
-
_DESCRIPTION = """\
|
32 |
-
CommonGen is a constrained text generation task, associated with a benchmark dataset,
|
33 |
-
to explicitly test machines for the ability of generative commonsense reasoning. Given
|
34 |
-
a set of common concepts; the task is to generate a coherent sentence describing an
|
35 |
-
everyday scenario using these concepts.
|
36 |
-
|
37 |
-
CommonGen is challenging because it inherently requires 1) relational reasoning using
|
38 |
-
background commonsense knowledge, and 2) compositional generalization ability to work
|
39 |
-
on unseen concept combinations. Our dataset, constructed through a combination of
|
40 |
-
crowd-sourcing from AMT and existing caption corpora, consists of 30k concept-sets and
|
41 |
-
50k sentences in total.
|
42 |
-
"""
|
43 |
-
_URL = "https://storage.googleapis.com/huggingface-nlp/datasets/common_gen/commongen_data.zip"
|
44 |
-
|
45 |
-
|
46 |
-
class CommonGen(datasets.GeneratorBasedBuilder):
|
47 |
-
VERSION = datasets.Version("2020.5.30")
|
48 |
-
|
49 |
-
def _info(self):
|
50 |
-
features = datasets.Features(
|
51 |
-
{
|
52 |
-
"concept_set_idx": datasets.Value("int32"),
|
53 |
-
"concepts": datasets.Sequence(datasets.Value("string")),
|
54 |
-
"target": datasets.Value("string"),
|
55 |
-
}
|
56 |
-
)
|
57 |
-
return datasets.DatasetInfo(
|
58 |
-
description=_DESCRIPTION,
|
59 |
-
features=features,
|
60 |
-
supervised_keys=datasets.info.SupervisedKeysData(input="concepts", output="target"),
|
61 |
-
homepage="https://inklab.usc.edu/CommonGen/index.html",
|
62 |
-
citation=_CITATION,
|
63 |
-
)
|
64 |
-
|
65 |
-
def _split_generators(self, dl_manager):
|
66 |
-
"""Returns SplitGenerators."""
|
67 |
-
|
68 |
-
dl_dir = dl_manager.download_and_extract(_URL)
|
69 |
-
|
70 |
-
return [
|
71 |
-
datasets.SplitGenerator(
|
72 |
-
name=datasets.Split.TRAIN,
|
73 |
-
gen_kwargs={"filepath": os.path.join(dl_dir, "commongen.train.jsonl"), "split": "train"},
|
74 |
-
),
|
75 |
-
datasets.SplitGenerator(
|
76 |
-
name=datasets.Split.VALIDATION,
|
77 |
-
gen_kwargs={"filepath": os.path.join(dl_dir, "commongen.dev.jsonl"), "split": "dev"},
|
78 |
-
),
|
79 |
-
datasets.SplitGenerator(
|
80 |
-
name=datasets.Split.TEST,
|
81 |
-
gen_kwargs={"filepath": os.path.join(dl_dir, "commongen.test_noref.jsonl"), "split": "test"},
|
82 |
-
),
|
83 |
-
]
|
84 |
-
|
85 |
-
def _generate_examples(self, filepath, split):
|
86 |
-
"""Yields examples."""
|
87 |
-
with open(filepath, encoding="utf-8") as f:
|
88 |
-
id_ = 0
|
89 |
-
for idx, row in enumerate(f):
|
90 |
-
row = row.replace(", }", "}") # Fix possible JSON format error
|
91 |
-
data = json.loads(row)
|
92 |
-
|
93 |
-
rand_order = [word for word in data["concept_set"].split("#")]
|
94 |
-
random.shuffle(rand_order)
|
95 |
-
|
96 |
-
if split == "test":
|
97 |
-
yield idx, {
|
98 |
-
"concept_set_idx": idx,
|
99 |
-
"concepts": rand_order,
|
100 |
-
"target": "",
|
101 |
-
}
|
102 |
-
else:
|
103 |
-
for scene in data["scene"]:
|
104 |
-
yield id_, {
|
105 |
-
"concept_set_idx": idx,
|
106 |
-
"concepts": rand_order,
|
107 |
-
"target": scene,
|
108 |
-
}
|
109 |
-
id_ += 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|