semeru/Text-Code-CodeSearchNet-Python

Dataset Overview

Source and Processing

• Source: The dataset originates from CodeSearchNet.
• Processing: Pre‑processed with CodeXGLUE scripts, which remove code samples that cannot be parsed into an abstract syntax tree and those that do not meet token count or language criteria.

Content

• Format: Three .jsonl files: train.jsonl, valid.jsonl, test.jsonl.
• Structure: Each line represents a function with fields such as repo, path, func_name, original_string, language, code/function, code_tokens/function_tokens, docstring, docstring_tokens, url, idx, etc.

Statistics

Task Definition

• Goal: Given a natural‑language description, retrieve matching source code.
• Test: Function names and variables in the test set are replaced with special tokens to evaluate generalization.

Example

• File: evaluator/test.jsonl
• Sample Content: Multiple records, each with url, docstring, function, idx, etc.

Prediction Input

• Processing: For each natural‑language query, rank candidate code snippets in descending order and return their idx values.
• Example Output: JSON format containing the url and an array of corresponding answers.