metadata

license: mit
Programminglanguage: Python
version: N/A
Date: Codesearchnet(Jun 2020 - paper release date)
Contaminated: Very Likely
Size: Standard Tokenizer (TreeSitter)

Dataset is imported from CodeXGLUE and pre-processed using their script.

Where to find in Semeru:

The dataset can be found at /nfs/semeru/semeru_datasets/code_xglue/text-to-code/codesearchnet/python in Semeru

CodeXGLUE -- Code Search (AdvTest)

Task Definition

Given a natural language, the task is to search source code that matches the natural language. To test the generalization ability of a model, function names and variables in test sets are replaced by special tokens.

Dataset

The dataset we use comes from CodeSearchNet and we filter the dataset as the following:

Remove examples that codes cannot be parsed into an abstract syntax tree.
Remove examples that #tokens of documents is < 3 or >256
Remove examples that documents contain special tokens (e.g. <img ...> or https:...)
Remove examples that documents are not English.

Besides, to test the generalization ability of a model, function names and variables in test sets are replaced by special tokens.

Data Format

After preprocessing dataset, you can obtain three .jsonl files, i.e. train.jsonl, valid.jsonl, test.jsonl

For each file, each line in the uncompressed file represents one function. One row is illustrated below.

repo: the owner/repo
path: the full path to the original file
func_name: the function or method name
original_string: the raw string before tokenization or parsing
language: the programming language
code/function: the part of the original_string that is code
code_tokens/function_tokens: tokenized version of code
docstring: the top-level comment or docstring, if it exists in the original string
docstring_tokens: tokenized version of docstring
url: the url for the example (identify natural language)
idx: the index of code (identify code)

Data Statistics

Data statistics of the dataset are shown in the below table:

	#Examples
Train	251,820
Dev	9,604
Test	19,210

Example

Given a text-code file evaluator/test.jsonl:

{"url": "url0", "docstring": "doc0","function": "fun0", "idx": 10}
{"url": "url1", "docstring": "doc1","function": "fun1", "idx": 11}
{"url": "url2", "docstring": "doc2","function": "fun2", "idx": 12}
{"url": "url3", "docstring": "doc3","function": "fun3", "idx": 13}
{"url": "url4", "docstring": "doc4","function": "fun4", "idx": 14}

Input Predictions

For each url for natural language, descending sort candidate codes and return their idx in order. For example:

{"url": "url0", "answers": [10,11,12,13,14]}
{"url": "url1", "answers": [10,12,11,13,14]}
{"url": "url2", "answers": [13,11,12,10,14]}
{"url": "url3", "answers": [10,14,12,13,11]}
{"url": "url4", "answers": [10,11,12,13,14]}

Reference

@article{husain2019codesearchnet,
  title={Codesearchnet challenge: Evaluating the state of semantic code search},
  author={Husain, Hamel and Wu, Ho-Hsiang and Gazit, Tiferet and Allamanis, Miltiadis and Brockschmidt, Marc},
  journal={arXiv preprint arXiv:1909.09436},
  year={2019}
}