HaofanWen/causal_debugging_agent

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Causal Debugging Agent

This project provides a causal debugging agent built on LangChain, LangGraph, and Gradio. It includes:

1. 1.Interactive Debugging Agent
2. 2.Perform causal analysis or code repair on input problems via a Gradio interface.

1. 1.Batch Generation Script
2. 2.Run the agent in batch mode to process a dataset of bugs.

1. 1.Evaluation Scripts
2. 2.Compute Exact Match, BLEU, and AST-based metrics on the generated fixes.

Requirements

• —Python 3.9 or higher
• —It is recommended to use a virtual environment (venv or conda)

Installation

# Clone the repository
git clone https://github.com/HaofanWen/causal_debugging_agent_cola_team.git
cd causal_debugging_agent_cola_team

# Install Python dependencies
pip install -r requirements.txt

Environment Variables

Create a .env file in the project root with the following keys:

# (Optional) Hugging Face username for Gradio sharing
HF_USERNAME=your_hf_username

# (Optional) Hugging Face Space ID
SPACE_ID=your_space_id

# Supabase database credentials
SUPABASE_URL=https://<your-project>.supabase.co
SUPABASE_SERVICE_KEY=your_service_key

Usage

1. Launch the Gradio Interface

python app.py

• —Go to the displayed URL in your browser.
• —Upload an output.json file and click Submit Uploaded Answers to view scores and answer tables.

2. Run Batch Generation

python main.py

• —By default, reads from bug_data/debug_dataset.jsonl and writes results to output.json.
• —To switch to the student dataset, uncomment and adjust the INPUT_PATH in main.py.

3. Evaluate Outputs

python debug_evaluate.py

• —Computes Exact Match, BLEU, and AST scores and saves them in evaluate_debug.json.

Project Structure

.
├── app.py                   # Gradio interface
├── main.py                  # Batch generation script
├── debug_evaluate.py        # Evaluation script
├── bug_data/                # Debug dataset (JSONL)
├── student_data/            # Student dataset (JSONL)
├── output.json              # Agent outputs
├── evaluate_debug.json      # Evaluation results
└── requirements.txt         # Python dependencies

License

This project is licensed under the Apache License 2.0.