LAMCode

An official source code for paper Language-Aware Code Representation Learning for Multilingual Code Understanding.

---

Requirements

The proposed LAMCode is implemented with python 3.7.11 on a NVIDIA Tesla V100 GPU.

• torch==1.5.0
• tqdm==4.62.2
• numpy==1.19.5
• scikit_learn==0.24.2

Quick Start

• Step1: data preparation

data link (mixed dataset):

NL Code Search:https://drive.google.com/file/d/1pV9-qQuSCk2OeA7efBuVd4EIRIhz_NAE/view?usp=sharing

XL Code Search:https://drive.google.com/file/d/1lqKIBoSLXXkIPAxzexvw9AwoD0jPsk0N/view?usp=sharing

other single data link:

https://github.com/reddy-lab-code-research/XLCoST

Download the data folders, unzip them, and put them directly in the 'nl2codesearch/dataset/program_level/' or 'code2codesearch/dataset/program_level/' directory.

• Step2: train

  bash run_code_search.sh 0 all nl2code program codebert train
  

• Step3: test

  bash run_code_search.sh 0 java code2code program codebert eval
  

Parameter setting

• GPU: 0
• Language: all
• task: nl2code
• model: codebert
• type: train