CodeDocAI: An NLP-Powered Code Documentation Generator

Project Overview

This project aims to help programmers write documentation for their code, using a fine-tuned T5 model to generate readable documentation from code chunks. The training is done on the command line and the documentation is generated on a web interface through Streamlit.

Features

• Documentation generation for code chunks.
• Interactive Web Interface made using Streamlit
• Model fine-tuned on code-documentation pairs.

Getting Started

1. Clone Repository

git clone https://github.com/Nebu0528/CodeDocAI

2. Install Dependencies

Run the command below to install all the dependencies needed:

pip install -r dependencies.txt

3. Train the Model

This step is necessary to teach the model how to interpret code and generate corresponding documentation based on that training. The model is being trained on this:

https://huggingface.co/datasets/jtatman/python-code-dataset-500k

• 1. Train the model using:

python src/model.py

• 2. After training, the model will be saved in the models/codetext_t5/ directory.

4. Run the Streamlit App

Once the model is trained, run the app:

streamlit run interface/website.py

5. Access App

Once the app is running, open the URL provided by Streamlit

http://localhost:8501

Usage

1. Paste a code snippet into the text area in the Streamlit interface.
2. Click "Generate Documentation" to generate the documentation for your code.

Example Code Snippet

def multiply(a, b):
    #This function returns the product of variables a and b
    return a * b

Future Improvements

1. Only works if you provide one function at a time with detailed comments
2. Integrate the Tranformers and BART Models to summarize the documentation
3. Handle different training dataset types (i.e .json, csv etc..)
4. Being able to download the generated documentation
5. Train the model on this dataset