GPhub-kit is a Python toolkit for benchmarking Gaussian Process (GP) libraries across multiple programming languages, enabling reproducible comparisons of implementations.

It provides a unified platform for:

• Multi-Language Execution: Run GP libraries in Python, R, Julia, and MATLAB from a unified interface
• Benchmarking: Selected standard engineering benchmarks
• Performance Metrics: Probabilistic metrics and timing/memory profiling
• Data Management: Synthetic data generation, dataset splitting

Quick Start

Command Line Interface

GPhub-kit provides a CLI for all benchmarking operations.

Project Management

# Create new project with directory structure
❯ gphubkit create --project PROJECT_NAME

# Add library implementations
❯ gphubkit add --name LIBRARY_NAME --language LANGUAGE
# Supported languages: Python, R, Julia, MATLAB

Running Benchmarks

# Standard benchmarks (BM01-BM07)
❯ gphubkit run bm --id {1-7}

# Composite Shell benchmark
❯ gphubkit run composite --dim {2,4,8,16,32,48,64} --size {100,200,500,700,1000,2000,3000}

# Custom benchmark with your data
❯ gphubkit run custom [--testsize FRACTION]

Post-processing

# Generate report and visualizations
❯ gphubkit postprocess

Python API

import gphubkit as gpk
from pathlib import Path

# 1. Load a benchmark - e.g. 2D Branin function
benchmark = gpk.benchmark.BM02()

# 2. Run benchmarking (requires library scripts in ./scripts/)
benchmark.run()

# 3. Postprocess results and generate reports
benchmark.postprocess()

Setup Project Structure

project/
├── scripts/                     # Library implementations
│   ├── lib_GPyTorch.py          # Python libraries
│   ├── lib_DiceKriging.r        # R libraries
│   ├── lib_GaussianProcesses.jl # Julia libraries
│   ├── lib_UQLab.m              # MATLAB libraries
│   └── ...                      # Additional libraries
├── data/                        # Benchmark data
│   ├── dataset_x.csv            # Dataset x csv file
│   ├── dataset_y.csv            # Dataset y csv file
│   ├── train_x.csv              # Train Set x csv file
│   ├── train_y.csv              # Train Set y csv file
│   ├── test_x.csv               # Test Set x csv file
│   └── test_y.csv               # Test Set y csv file
└── results/                     # Rich console output
    ├── storage/                 # Parquet result files
    │   └── ...
    ├── img/                     # Plots and visualizations
    │   └── ...
    ├── logger.log               # Execution log
    └── report.log               # Analysis report

---

Building and Running

Installation

The easiest way to install GPhub-kit is using uv. The main dependencies are listed in pyproject.toml and include libraries for GP modeling, data handling, and plotting.

1. Create a virtual environment

uv venv --python X.XX.X

2. Activate the virtual environment (macOS/Linux)

source .venv/bin/activate

3. Install the dependencies

uv add gphubkit

4. Install additional dependencies

# For GPy and SMT support
uv add "gphubkit[gpy,smt]"
# or for all optional dependencies
uv add "gphubkit[all]"

Requirements

MATLAB, Julia, and R must be installed on your system. The gphubkit package interfaces with these languages through matlabengine, r2py, and juliacall respectively. Please refer to each library's documentation for installation instructions.

---

Documentation

Documentation for all modules is provided in the 📚 Wiki of the repo:

• Benchmark Module: Standard and custom benchmarks
• Data Module: Dataset management, synthetic generation, file I/O
• Metrics Module: Performance metrics and reporting
• Launcher Module: Cross-language execution engines and pipeline
• Plotter Module: Visualization suite
• Utils Module: Internal utilities and helpers

---

Development

To clone and set up the development environment:

git clone https://github.com/alefaraci/GPhub-kit.git
cd GPhub-kit
uv venv --python X.XX.X
source .venv/bin/activate  # macOS/Linux
uv add --dev "gphubkit[all]"

# Test CLI installation
gphubkit --help

Contributing

GPhub-kit is designed to be extensible and welcomes contributions:

• New Benchmarks: Add domain-specific benchmark problems
• Language Support: Extend support to additional programming languages
• Metrics: Implement new performance and statistical metrics
• Visualization: Create specialized plotting functions for specific analyses
• Documentation: Improve examples and use-case documentation

---

Acknowledgments

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 955393.

License

Licensed under the MIT license.