shuyanzh commited on
Commit
297d935
1 Parent(s): 847a25a

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +95 -0
README.md ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ annotations_creators: []
3
+ language_creators:
4
+ - crowdsourced
5
+ - expert-generated
6
+ language:
7
+ - code
8
+ license:
9
+ - mit
10
+ multilinguality:
11
+ - monolingual
12
+ size_categories:
13
+ - unknown
14
+ source_datasets:
15
+ - original
16
+ task_categories:
17
+ - text2text-generation
18
+ task_ids: []
19
+ pretty_name: DocPrompting-CoNaLa
20
+ tags:
21
+ - code-generation
22
+ - doc retrieval
23
+ - retrieval augmented generation
24
+ ---
25
+
26
+ ## Dataset Description
27
+ - **Repository:** https://github.com/shuyanzhou/docprompting
28
+ - **Paper:** [DocPrompting: Generating Code by Retrieving the Docs](https://arxiv.org/pdf/2207.05987.pdf)
29
+
30
+ ### Dataset Summary
31
+ This is the re-split of [CoNaLa](https://conala-corpus.github.io/) dataset.
32
+ For each code snippet in the dev and test set, at least one function is held out from the training set.
33
+ This split aims at testing a code generation model's capacity in generating *unseen* functions
34
+ We further make sure that examples from the same StackOverflow post (same `question_id` before `-`) are in the same split.
35
+
36
+ ### Supported Tasks and Leaderboards
37
+ This dataset is used to evaluate code generations.
38
+
39
+ ### Languages
40
+ English - Python code.
41
+
42
+ ## Dataset Structure
43
+ ```python
44
+ dataset = load_dataset("neulab/docpromting-conala")
45
+ DatasetDict({
46
+ train: Dataset({
47
+ features: ['nl', 'cmd', 'question_id', 'cmd_name', 'oracle_man', 'canonical_cmd'],
48
+ num_rows: 2135
49
+ })
50
+ test: Dataset({
51
+ features: ['nl', 'cmd', 'question_id', 'cmd_name', 'oracle_man', 'canonical_cmd'],
52
+ num_rows: 543
53
+ })
54
+ validation: Dataset({
55
+ features: ['nl', 'cmd', 'question_id', 'cmd_name', 'oracle_man', 'canonical_cmd'],
56
+ num_rows: 201
57
+ })
58
+ })
59
+ })
60
+
61
+ code_docs = load_dataset("neulab/docprompting-conala", "docs")
62
+ DatasetDict({
63
+ train: Dataset({
64
+ features: ['doc_id', 'doc_content'],
65
+ num_rows: 34003
66
+ })
67
+ })
68
+ ```
69
+
70
+ ### Data Fields
71
+ train/dev/test:
72
+ - nl: The natural language intent
73
+ - cmd: The reference code snippet
74
+ - question_id: `x-y`where `x` is the StackOverflow post ID
75
+ - oracle_man: The `doc_id` of the functions used in the reference code snippet. The corresponding contents are in `doc` split
76
+ - canonical_cmd: The canonical version reference code snippet
77
+
78
+
79
+ docs:
80
+ - doc_id: the id of a doc
81
+ - doc_content: the content of the doc
82
+
83
+ ## Dataset Creation
84
+ The dataset was crawled from Stack Overflow, automatically filtered, then curated by annotators. For more details, please refer to the original [paper](https://arxiv.org/pdf/1805.08949.pdf)
85
+
86
+ ### Citation Information
87
+
88
+ ```
89
+ @article{zhou2022doccoder,
90
+ title={DocCoder: Generating Code by Retrieving and Reading Docs},
91
+ author={Zhou, Shuyan and Alon, Uri and Xu, Frank F and JIang, Zhengbao and Neubig, Graham},
92
+ journal={arXiv preprint arXiv:2207.05987},
93
+ year={2022}
94
+ }
95
+ ```