dimreduce4gpu

dimreduce4gpu is a GPU-accelerated dimensionality reduction library built with CUDA, designed for fast and efficient large-scale data reduction. It provides implementations of popular algorithms like Principal Component Analysis (PCA) and Truncated Singular Value Decomposition (SVD), optimized to harness GPU power—making it ideal for high-performance applications in data science and machine learning.

---

🚀 Features

• GPU-Accelerated: Leverages CUDA to achieve significant speedups on large datasets.
• Optimized Implementations: Includes PCA and Truncated SVD tailored for high throughput and scale.
• Python Integration: Easily integrates into Python-based data workflows.

✅ Modern builds and CI

• CPU-only installs are supported via a native C++ backend (libdimreduce4cpu.*).
• GPU acceleration uses the CUDA backend (libdimreduce4gpu.*) when available.
• GitHub Actions runs unit tests on CPU runners, and includes a build+verify job for the native libraries.
• A dedicated workflow builds manylinux CPU wheels: .github/workflows/wheels.yml.

Backend selection

Both PCA and TruncatedSVD accept backend:

• backend="auto" (default): GPU if runnable, else CPU
• backend="cpu": force CPU backend
• backend="gpu": force GPU backend

---

📌 Supported Algorithms

• Principal Component Analysis (PCA)
  Reduces dimensionality by transforming variables into a set of linearly uncorrelated principal components.

• Truncated Singular Value Decomposition (SVD)
  Approximates SVD by retaining only the most significant singular values, making it suitable for sparse and large-scale datasets.

---

🛠 Build Instructions

📋 Requirements

• Python: 3.9+
• Build tools: CMake 3.18+, a C++17 compiler
• CPU backend: BLAS + LAPACK development headers (e.g., OpenBLAS)
• GPU backend (optional): CUDA toolkit + NVIDIA driver/runtime

Quickstart (CPU)

python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install .
pytest -q

Building the native libraries (developers)

CPU-only build:

cmake -S . -B build/cpu -DCMAKE_BUILD_TYPE=Release -DDIMREDUCE4GPU_BUILD_CPU=ON -DDIMREDUCE4GPU_BUILD_CUDA=OFF
cmake --build build/cpu -j

CUDA build (requires CUDA toolkit):

cmake -S . -B build/cuda -DCMAKE_BUILD_TYPE=Release -DDIMREDUCE4GPU_BUILD_CPU=ON -DDIMREDUCE4GPU_BUILD_CUDA=ON
cmake --build build/cuda -j

📦 Integration in Other Projects

dimreduce4gpu is also part of other GPU-optimized machine learning ecosystems:

• H2O4GPU by H2O.ai
  • 🔹 Truncated SVD Module
  • 🔹 PCA Module

---

🤝 Contributing

We welcome contributions! Feel free to:

• 🐛 Open an issue for bugs or feature requests
• 💬 Ask questions or share ideas
• 🔧 Submit pull requests to improve the project

Thank you for using dimreduce4gpu!

CPU backend implementation

See docs/CPU_BACKEND.md for a detailed explanation of the CPU PCA/TruncatedSVD algorithms and how parity is tested against scikit-learn.

Benchmarks

See docs/BENCHMARKS.md and bench/benchmark_cpu_vs_sklearn.py for CPU performance comparisons against scikit-learn.