Generative Nonparametric Conditional Density Estimation

A benchmark comparing traditional statistical methods and generative modeling approaches for conditional density estimation.

Installation

Quick Setup

# Clone the repository
git clone https://github.com/chiatsewang/generative-nonparametric-cde.git
cd generative-nonparametric-cde

# Install the package
pip install -e .

# Install development dependencies (optional)
pip install -e ".[dev]"

# Install FlexCode (optional, for FlexCode experiments)
pip install -e ".[flexcode]"

# Setup pre-commit hooks (optional)
pre-commit install

Requirements

• Python >= 3.11
• PyTorch >= 1.11.0
• NumPy, SciPy, matplotlib
• See pyproject.toml for full dependency list

Available Methods

• HyAlg - Kernel-based method (Hall & Yao method)
• FlexCode - Flexible conditional density estimation
• DeepCDE - Deep Conditional Density Estimation
• GCDS - Generative Conditional Density Sampling
• DDPM - Denoising Diffusion Probabilistic Models

Project Structure

.
├── condgen_benchmark/    # Main package
│   ├── data/             # Data loading and generation
│   ├── methods/          # CDE methods (ddpm, deepcde, gcds, hyalg)
│   └── models/           # Neural network architectures
├── scripts/              # Training and evaluation scripts
├── datasets/             # Generated datasets
└── workspaces/           # Model checkpoints and results

License

GPL-2.0 - see LICENSE