Upload odex.py

#1
by loubnabnl HF staff - opened
Files changed (2) hide show
  1. README.md +6 -0
  2. odex.py +82 -0
README.md CHANGED
@@ -16,6 +16,12 @@ __ODEX__ is an Open-Domain EXecution-based NL-to-Code generation data benchmark.
16
  It contains 945 samples with a total of 1,707 human-written test cases, covering intents in four different natural languages -- 439 in English, 90 in Spanish, 164 in Japanese, and 252 in Russian.
17
 
18
 
 
 
 
 
 
 
19
 
20
  If you find our dataset useful, please cite the paper
21
 
 
16
  It contains 945 samples with a total of 1,707 human-written test cases, covering intents in four different natural languages -- 439 in English, 90 in Spanish, 164 in Japanese, and 252 in Russian.
17
 
18
 
19
+ You can load the dataset by specifying a subset from *en, es, ja, ru* (by default the english subset *en* is loaded):
20
+ ```python
21
+ from datasets import load_dataset
22
+
23
+ ds = load_dataset("neulab/odex", "ja", split="test")
24
+ ```
25
 
26
  If you find our dataset useful, please cite the paper
27
 
odex.py ADDED
@@ -0,0 +1,82 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Odex dataset."""
2
+
3
+ import json
4
+ import datasets
5
+
6
+
7
+ _CITATION = """\
8
+ @article{wang2022execution,
9
+ title={Execution-Based Evaluation for Open-Domain Code Generation},
10
+ author={Wang, Zhiruo and Zhou, Shuyan and Fried, Daniel and Neubig, Graham},
11
+ journal={arXiv preprint arXiv:2212.10481},
12
+ year={2022}
13
+ }
14
+ """
15
+
16
+ _DESCRIPTION = """\
17
+ ODEX is an Open-Domain EXecution-based NL-to-Code generation data benchmark.
18
+ It contains 945 samples with a total of 1,707 human-written test cases,
19
+ covering intents in four different natural languages -- 439 in English, 90 in Spanish, 164 in Japanese, and 252 in Russian.
20
+ """
21
+
22
+ _HOMEPAGE = "https://github.com/zorazrw/odex"
23
+ _URLs = {
24
+ "en": "en_test.jsonl",
25
+ "es": "es_test.jsonl",
26
+ "ja": "ja_test.jsonl",
27
+ "ru": "ru_test.jsonl",
28
+ }
29
+
30
+ class Odex(datasets.GeneratorBasedBuilder):
31
+ """ODEX Code dataset."""
32
+
33
+ VERSION = datasets.Version("1.0.0")
34
+
35
+
36
+ BUILDER_CONFIGS = [
37
+ datasets.BuilderConfig(
38
+ name=lang,
39
+ version=datasets.Version("1.0.0"),
40
+ description=_DESCRIPTION,
41
+ ) for lang in _URLs.keys()
42
+ ]
43
+
44
+ DEFAULT_CONFIG_NAME = "en"
45
+
46
+
47
+ def _info(self):
48
+ features = datasets.Features({"task_id": datasets.Value("int64"),
49
+ "prompt": datasets.Value("string"),
50
+ "suffix": datasets.Value("string"),
51
+ "canonical_solution": datasets.Value("string"),
52
+ "test_start": datasets.Value("string"),
53
+ "test": datasets.Sequence(datasets.Value("string")),
54
+ "entry_point": datasets.Value("string"),
55
+ "intent": datasets.Value("string"),
56
+ "library": datasets.Sequence(datasets.Value("string")),
57
+ })
58
+ return datasets.DatasetInfo(
59
+ description=_DESCRIPTION,
60
+ features=features,
61
+ supervised_keys=None,
62
+ citation=_CITATION,
63
+ homepage=_HOMEPAGE)
64
+
65
+ def _split_generators(self, dl_manager):
66
+ """Returns SplitGenerators."""
67
+ config_urls = _URLs[self.config.name]
68
+ data_dir = dl_manager.download_and_extract(config_urls)
69
+ return [
70
+ datasets.SplitGenerator(
71
+ name=datasets.Split.TEST,
72
+ gen_kwargs={"filepath": data_dir, "split": "train"},
73
+ ),
74
+ ]
75
+
76
+
77
+ def _generate_examples(self, filepath, split):
78
+ key = 0
79
+ for line in open(filepath, encoding="utf-8"):
80
+ line = json.loads(line)
81
+ yield key, line
82
+ key += 1