# PaCMAP-GPU: GPU-Accelerated Pairwise Controlled Manifold Approximation
[](https://www.python.org/)
[](https://developer.nvidia.com/cuda-toolkit)
[](https://pytorch.org/)
[](https://rapids.ai/)
[](https://opensource.org/licenses/Apache-2.0)
**PaCMAP-GPU** is a high-performance GPU implementation of Pairwise Controlled Manifold Approximation and Projection (PaCMAP), providing dramatic speedups for large-scale dimensionality reduction and manifold learning tasks.
## 🚀 Key Features
- **GPU Acceleration**: Native CUDA implementation using PyTorch and CUDA graphs for maximum performance
- **Massive Speedups**: 5-50x faster than CPU implementations depending on dataset size
- **Memory Efficient**: Optimized GPU memory usage with automatic cleanup
- **Drop-in Replacement**: Compatible API with sklearn-style transformers
- **Three-Phase Training**: Standard PaCMAP phases plus LocalMAP variant
- **Large Dataset Support**: Handle datasets with millions of points efficiently
- **CUDA Graph Optimization**: Minimal kernel launch overhead
## 📋 Requirements
### Hardware Requirements
- NVIDIA GPU with Compute Capability 6.0+ (Pascal architecture or newer)
- 4GB+ GPU memory (8GB+ recommended for large datasets)
- CUDA 12.0 or higher
### Software Dependencies
Core dependencies (automatically installed):
- PyTorch >= 2.0.0
- NumPy >= 1.21.0, < 2.0.0
- SciPy >= 1.7.0
- scikit-learn >= 1.0.0
- pandas >= 1.3.0
- matplotlib >= 3.5.0
GPU acceleration (required):
- cupy-cuda12x >= 12.0.0
- cuml-cu12 >= 24.0.0
- cudf-cu12 >= 24.0.0
- faiss-gpu >= 1.7.0
**Important: CUDA-enabled PyTorch Required**
> By default, `pip install pacmap-gpu` will install the CPU-only version of PyTorch.
> **To use GPU acceleration, you must manually install the CUDA-enabled PyTorch wheel** that matches your system and CUDA version.
> See [PyTorch Get Started](https://pytorch.org/get-started/locally/) for installation instructions.
**Example:**
```bash
# Example for CUDA 12.1
pip install torch==2.2.2+cu121 torchvision==0.17.2+cu121 torchaudio==2.2.2+cu121 \
--index-url https://download.pytorch.org/whl/cu121
```
## 🔧 Installation
### Option 1: Install from PyPI (Recommended)
```bash
# Install PaCMAP-GPU with all dependencies
pip install pacmap-gpu
# Verify installation
pacmap-gpu-check
```
**Note:** PaCMAP-GPU requires CUDA-capable hardware and GPU libraries. All necessary RAPIDS and CUDA dependencies are automatically installed.
### Option 2: Install from Source
```bash
# Clone the repository
git clone https://github.com/AntonioRoye/PaCMAP-GPU.git
cd PaCMAP-GPU
# Create conda environment with RAPIDS and all dependencies (recommended for development)
conda create -n pacmap-gpu -c rapidsai -c conda-forge \
cuml-cu12 cudf-cu12 cupy-cuda12x faiss-gpu \
python=3.10 pytorch pytorch-cuda=12.1 -c pytorch -c nvidia
# Activate environment
conda activate pacmap-gpu
# Install package in development mode
pip install -e .
```
### Alternative: Using requirements file
```bash
# Clone repository
git clone https://github.com/AntonioRoye/PaCMAP-GPU.git
cd PaCMAP-GPU
# Create virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install CUDA 12 dependencies
pip install -r requirements-cuda12.txt
# Install package
pip install -e .
```
### Verify Installation
```bash
# Check installation
pacmap-gpu-check
# Or in Python
python -c "from pacmap_gpu import PaCMAPGPU, check_gpu_availability; print('GPU Available:', check_gpu_availability())"
```
## 🎯 Quick Start
### Basic Usage
```python
import numpy as np
from pacmap_gpu import PaCMAPGPU
from sklearn.datasets import make_swiss_roll
# Generate sample data
X, color = make_swiss_roll(n_samples=2000, noise=0.1, random_state=42)
# Create PaCMAP-GPU instance
reducer = PaCMAPGPU(
n_components=2,
n_neighbors=10,
num_iters=450,
random_state=42,
verbose=True
)
# Fit and transform the data
X_embedded = reducer.fit_transform(X)
print(f"Embedded shape: {X_embedded.shape}")
```
### Advanced Configuration
```python
# High-performance configuration for large datasets
reducer = PaCMAPGPU(
n_components=2,
n_neighbors=15,
MN_ratio=0.5,
FP_ratio=2.0,
num_iters=450,
batch_size=8192,
init_method="pca",
optimizer="adam",
lr=0.01,
low_dist_thres=1.0, # LocalMAP parameter
device="cuda",
verbose=True,
random_state=42
)
X_embedded = reducer.fit_transform(X)
```
### Visualization Example
```python
import matplotlib.pyplot as plt
from sklearn.datasets import load_digits
# Load digits dataset
digits = load_digits()
X, y = digits.data, digits.target
# Apply PaCMAP-GPU
reducer = PaCMAPGPU(n_components=2, verbose=True, random_state=42)
X_embedded = reducer.fit_transform(X)
# Plot results
plt.figure(figsize=(10, 8))
scatter = plt.scatter(X_embedded[:, 0], X_embedded[:, 1],
c=y, cmap='tab10', alpha=0.7)
plt.colorbar(scatter)
plt.title('Digits Dataset - PaCMAP-GPU Embedding')
plt.xlabel('Component 1')
plt.ylabel('Component 2')
plt.show()
```
## 📊 Performance Benchmarks
### Speedup Comparison
Performance varies based on dataset size and GPU hardware. Below are typical dataset configurations for benchmarking:
| Dataset Size | Features | Use Case | Notes |
|-------------|----------|----------|-------|
| 1K samples | 50 | Quick testing | May show minimal speedup due to GPU overhead |
| 5K samples | 100 | Standard evaluation | ~2-5x speedup typical |
| 10K samples | 200 | Medium datasets | ~5-15x speedup expected |
| 50K samples | 500 | Large-scale analysis | ~10-50x speedup possible |
*Actual performance depends on GPU model, CUDA version, and data characteristics*
**Run your own benchmarks:**
```bash
pytest tests/test_benchmarks.py -v -m benchmark
```
### Memory Efficiency
- **PaCMAP-GPU**: ~2-4GB GPU memory for 50K samples (with CUDA graph optimizations)
- **Traditional Methods**: ~4-8GB RAM for 50K samples (varies by dimensionality)
- **Automatic Cleanup**: Memory pools and CUDA graph optimization prevent OOM errors
## 🔍 API Reference
### PaCMAPGPU Class
```python
PaCMAPGPU(
n_components=2, # Output dimensions
n_neighbors=10, # Neighbors for local structure
MN_ratio=0.5, # Mid-near pairs ratio
FP_ratio=2.0, # Further pairs ratio
num_iters=450, # Total optimization iterations
batch_size=8192, # Batch size for stochastic optimization
init_method="pca", # Initialization method
optimizer="adam", # Optimizer type
lr=0.01, # Learning rate
device="cuda", # CUDA device
low_dist_thres=1.0, # LocalMAP distance threshold
random_state=None, # Random seed
verbose=False # Print progress
)
```
#### Key Methods
- `fit_transform(X)`: Fit and return embedding
- `transform(X)`: Transform new data (not implemented)
#### Attributes
- `embedding_`: Final embedding coordinates
- `is_fitted`: Whether model has been fitted
- `n_samples`: Number of training samples
### Utility Functions
```python
from pacmap_gpu import check_gpu_availability, get_gpu_info, clear_gpu_memory
# Check GPU status
gpu_available = check_gpu_availability()
gpu_info = get_gpu_info()
# Memory management
clear_gpu_memory()
# Installation verification
from pacmap_gpu import verify_installation, print_installation_report
print_installation_report()
```
## ⚡ Performance Tips
### Dataset Size Guidelines
- **Small datasets (< 1K)**: GPU overhead may impact performance, but still required
- **Medium datasets (1K-10K)**: GPU shows significant speedup over traditional CPU methods
- **Large datasets (10K+)**: Maximum GPU performance advantage
### Memory Optimization
```python
from pacmap_gpu import clear_gpu_memory
# Clear memory between runs
clear_gpu_memory()
# Use efficient data types
X = X.astype(np.float32) # Use float32 instead of float64
```
### Hardware Optimization
- Use modern GPUs with Tensor Cores (RTX series, V100, A100)
- Ensure sufficient GPU memory (8GB+ for datasets > 50K samples)
- Use CUDA 12.0+ for best performance
## 🧪 Running Tests and Benchmarks
```bash
# Install development dependencies
pip install -e ".[dev]"
# Run basic tests
pytest tests/ -v
# Run GPU tests (requires GPU)
pytest tests/ -v -m gpu
# Run benchmarks
pytest tests/test_benchmarks.py -v -m benchmark
# Run without slow tests
pytest tests/ -v -m "not slow"
```
### Example Benchmark Output
```bash
pytest tests/test_benchmarks.py::TestPerformanceBenchmarks::test_gpu_vs_cpu_performance -v -s
=== MEDIUM Dataset Benchmark ===
Dataset shape: (2000, 50)
GPU time: 1.8432s
CPU time: 14.2156s
Speedup: 7.71x
```
## 🐛 Troubleshooting
### Common Issues
1. **CUDA Not Found**
```bash
# Check CUDA installation
nvidia-smi
python -c "import torch; print(torch.cuda.is_available())"
```
2. **Out of Memory Error**
```python
# Reduce batch size or data size
reducer = PaCMAPGPU(batch_size=2048, n_neighbors=5)
```
3. **Import Errors**
```bash
# Verify installation
pacmap-gpu-check
# Reinstall if needed
pip install --force-reinstall pacmap-gpu
```
### Performance Issues
- **Slow on small datasets**: GPU overhead exists, but GPU is still required for functionality
- **GPU underutilized**: Increase dataset size or batch parameters
- **Memory warnings**: Monitor GPU memory and clear caches regularly
## 🤝 Contributing
We welcome contributions! Please follow these steps:
1. Fork the repository
2. Create a feature branch (`git checkout -b feature/amazing-feature`)
3. Make changes and add tests
4. Run the test suite (`pytest tests/`)
5. Commit changes (`git commit -m 'Add amazing feature'`)
6. Push to branch (`git push origin feature/amazing-feature`)
7. Open a Pull Request
### Development Setup
```bash
# Clone repository
git clone https://github.com/AntonioRoye/PaCMAP-GPU.git
cd PaCMAP-GPU
# Set up conda environment with all dependencies (recommended)
conda create -n pacmap-gpu-dev -c rapidsai -c conda-forge \
cuml-cu12 cudf-cu12 cupy-cuda12x faiss-gpu \
python=3.10 pytorch pytorch-cuda=12.1 -c pytorch -c nvidia
conda activate pacmap-gpu-dev
# Install package and development tools
pip install -e ".[dev]"
# Or alternatively, install from requirements file + dev extras
# pip install -r requirements-cuda12.txt
# pip install -e ".[dev]"
# Run tests
pytest tests/
# Run GPU-specific tests (requires CUDA)
pytest tests/ -m gpu
# Format code
black source/
# Type checking
mypy source/
```
## 📚 Citation
If you use PaCMAP-GPU in your research, please cite:
```bibtex
@article{JMLR:v22:20-1061,
author = {Yingfan Wang and Haiyang Huang and Cynthia Rudin and Yaron Shaposhnik},
title = {Understanding How Dimension Reduction Tools Work: An Empirical Approach to Deciphering t-SNE, UMAP, TriMap, and PaCMAP for Data Visualization},
journal = {Journal of Machine Learning Research},
year = {2021},
volume = {22},
number = {201},
pages = {1-73},
url = {http://jmlr.org/papers/v22/20-1061.html}
}
@software{pacmap_gpu2025,
title={PaCMAP-GPU: GPU-Accelerated Pairwise Controlled Manifold Approximation},
author={Antonio Roye-Azar},
year={2025},
url={https://github.com/AntonioRoye/PaCMAP-GPU}
}
```
## 📄 License
This project is licensed under the Apache License 2.0 - see the [LICENSE](LICENSE) file for details.
## 🔗 Links
- **Original PaCMAP**: [https://github.com/YingfanWang/PaCMAP](https://github.com/YingfanWang/PaCMAP)
- **PyTorch**: [https://pytorch.org/](https://pytorch.org/)
- **RAPIDS cuML**: [https://docs.rapids.ai/api/cuml/stable/](https://docs.rapids.ai/api/cuml/stable/)
- **Issues**: [https://github.com/AntonioRoye/PaCMAP-GPU/issues](https://github.com/AntonioRoye/PaCMAP-GPU/issues)
## 🙏 Acknowledgments
- Original PaCMAP authors for the algorithm and reference implementation
- RAPIDS team for GPU acceleration libraries
- PyTorch team for the deep learning framework
- NVIDIA for CUDA and GPU computing platform
---
**Made with ❤️ for the GPU computing community**
Raw data
{
"_id": null,
"home_page": "https://github.com/AntonioRoye/PaCMAP-GPU",
"name": "pacmap-gpu",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.10",
"maintainer_email": null,
"keywords": "dimensionality-reduction, manifold-learning, gpu, cuda, machine-learning",
"author": "Antonio Roye-Azar",
"author_email": "Antonio Roye-Azar <antonio@varonova.ca>",
"download_url": "https://files.pythonhosted.org/packages/2a/b9/c49ee9c570631eac95ec89cb362eea11460d36befe000282edf2e92e79fb/pacmap_gpu-1.0.2.tar.gz",
"platform": "Linux",
"description": "# PaCMAP-GPU: GPU-Accelerated Pairwise Controlled Manifold Approximation\n\n[](https://www.python.org/)\n[](https://developer.nvidia.com/cuda-toolkit)\n[](https://pytorch.org/)\n[](https://rapids.ai/)\n[](https://opensource.org/licenses/Apache-2.0)\n\n**PaCMAP-GPU** is a high-performance GPU implementation of Pairwise Controlled Manifold Approximation and Projection (PaCMAP), providing dramatic speedups for large-scale dimensionality reduction and manifold learning tasks.\n\n## \ud83d\ude80 Key Features\n\n- **GPU Acceleration**: Native CUDA implementation using PyTorch and CUDA graphs for maximum performance\n- **Massive Speedups**: 5-50x faster than CPU implementations depending on dataset size\n- **Memory Efficient**: Optimized GPU memory usage with automatic cleanup\n- **Drop-in Replacement**: Compatible API with sklearn-style transformers\n- **Three-Phase Training**: Standard PaCMAP phases plus LocalMAP variant\n- **Large Dataset Support**: Handle datasets with millions of points efficiently\n- **CUDA Graph Optimization**: Minimal kernel launch overhead\n\n## \ud83d\udccb Requirements\n\n### Hardware Requirements\n- NVIDIA GPU with Compute Capability 6.0+ (Pascal architecture or newer)\n- 4GB+ GPU memory (8GB+ recommended for large datasets)\n- CUDA 12.0 or higher\n\n### Software Dependencies\nCore dependencies (automatically installed):\n- PyTorch >= 2.0.0\n- NumPy >= 1.21.0, < 2.0.0\n- SciPy >= 1.7.0\n- scikit-learn >= 1.0.0\n- pandas >= 1.3.0\n- matplotlib >= 3.5.0\n\nGPU acceleration (required):\n- cupy-cuda12x >= 12.0.0\n- cuml-cu12 >= 24.0.0\n- cudf-cu12 >= 24.0.0\n- faiss-gpu >= 1.7.0\n\n**Important: CUDA-enabled PyTorch Required**\n\n> By default, `pip install pacmap-gpu` will install the CPU-only version of PyTorch. \n> **To use GPU acceleration, you must manually install the CUDA-enabled PyTorch wheel** that matches your system and CUDA version. \n> See [PyTorch Get Started](https://pytorch.org/get-started/locally/) for installation instructions.\n\n**Example:**\n```bash\n# Example for CUDA 12.1\npip install torch==2.2.2+cu121 torchvision==0.17.2+cu121 torchaudio==2.2.2+cu121 \\\n --index-url https://download.pytorch.org/whl/cu121\n```\n\n## \ud83d\udd27 Installation\n\n### Option 1: Install from PyPI (Recommended)\n\n```bash\n# Install PaCMAP-GPU with all dependencies\npip install pacmap-gpu\n\n# Verify installation\npacmap-gpu-check\n```\n\n**Note:** PaCMAP-GPU requires CUDA-capable hardware and GPU libraries. All necessary RAPIDS and CUDA dependencies are automatically installed.\n\n### Option 2: Install from Source\n\n```bash\n# Clone the repository\ngit clone https://github.com/AntonioRoye/PaCMAP-GPU.git\ncd PaCMAP-GPU\n\n# Create conda environment with RAPIDS and all dependencies (recommended for development)\nconda create -n pacmap-gpu -c rapidsai -c conda-forge \\\n cuml-cu12 cudf-cu12 cupy-cuda12x faiss-gpu \\\n python=3.10 pytorch pytorch-cuda=12.1 -c pytorch -c nvidia\n\n# Activate environment\nconda activate pacmap-gpu\n\n# Install package in development mode\npip install -e .\n```\n\n### Alternative: Using requirements file\n\n```bash\n# Clone repository\ngit clone https://github.com/AntonioRoye/PaCMAP-GPU.git\ncd PaCMAP-GPU\n\n# Create virtual environment\npython -m venv venv\nsource venv/bin/activate # On Windows: venv\\Scripts\\activate\n\n# Install CUDA 12 dependencies\npip install -r requirements-cuda12.txt\n\n# Install package\npip install -e .\n```\n\n### Verify Installation\n\n```bash\n# Check installation\npacmap-gpu-check\n\n# Or in Python\npython -c \"from pacmap_gpu import PaCMAPGPU, check_gpu_availability; print('GPU Available:', check_gpu_availability())\"\n```\n\n## \ud83c\udfaf Quick Start\n\n### Basic Usage\n\n```python\nimport numpy as np\nfrom pacmap_gpu import PaCMAPGPU\nfrom sklearn.datasets import make_swiss_roll\n\n# Generate sample data\nX, color = make_swiss_roll(n_samples=2000, noise=0.1, random_state=42)\n\n# Create PaCMAP-GPU instance\nreducer = PaCMAPGPU(\n n_components=2,\n n_neighbors=10,\n num_iters=450,\n random_state=42,\n verbose=True\n)\n\n# Fit and transform the data\nX_embedded = reducer.fit_transform(X)\nprint(f\"Embedded shape: {X_embedded.shape}\")\n```\n\n### Advanced Configuration\n\n```python\n# High-performance configuration for large datasets\nreducer = PaCMAPGPU(\n n_components=2,\n n_neighbors=15,\n MN_ratio=0.5,\n FP_ratio=2.0,\n num_iters=450,\n batch_size=8192,\n init_method=\"pca\",\n optimizer=\"adam\",\n lr=0.01,\n low_dist_thres=1.0, # LocalMAP parameter\n device=\"cuda\",\n verbose=True,\n random_state=42\n)\n\nX_embedded = reducer.fit_transform(X)\n```\n\n### Visualization Example\n\n```python\nimport matplotlib.pyplot as plt\nfrom sklearn.datasets import load_digits\n\n# Load digits dataset\ndigits = load_digits()\nX, y = digits.data, digits.target\n\n# Apply PaCMAP-GPU\nreducer = PaCMAPGPU(n_components=2, verbose=True, random_state=42)\nX_embedded = reducer.fit_transform(X)\n\n# Plot results\nplt.figure(figsize=(10, 8))\nscatter = plt.scatter(X_embedded[:, 0], X_embedded[:, 1], \n c=y, cmap='tab10', alpha=0.7)\nplt.colorbar(scatter)\nplt.title('Digits Dataset - PaCMAP-GPU Embedding')\nplt.xlabel('Component 1')\nplt.ylabel('Component 2')\nplt.show()\n```\n\n## \ud83d\udcca Performance Benchmarks\n\n### Speedup Comparison\n\nPerformance varies based on dataset size and GPU hardware. Below are typical dataset configurations for benchmarking:\n\n| Dataset Size | Features | Use Case | Notes |\n|-------------|----------|----------|-------|\n| 1K samples | 50 | Quick testing | May show minimal speedup due to GPU overhead |\n| 5K samples | 100 | Standard evaluation | ~2-5x speedup typical | \n| 10K samples | 200 | Medium datasets | ~5-15x speedup expected |\n| 50K samples | 500 | Large-scale analysis | ~10-50x speedup possible |\n\n*Actual performance depends on GPU model, CUDA version, and data characteristics*\n\n**Run your own benchmarks:**\n```bash\npytest tests/test_benchmarks.py -v -m benchmark\n```\n\n### Memory Efficiency\n\n- **PaCMAP-GPU**: ~2-4GB GPU memory for 50K samples (with CUDA graph optimizations)\n- **Traditional Methods**: ~4-8GB RAM for 50K samples (varies by dimensionality) \n- **Automatic Cleanup**: Memory pools and CUDA graph optimization prevent OOM errors\n\n## \ud83d\udd0d API Reference\n\n### PaCMAPGPU Class\n\n```python\nPaCMAPGPU(\n n_components=2, # Output dimensions\n n_neighbors=10, # Neighbors for local structure\n MN_ratio=0.5, # Mid-near pairs ratio\n FP_ratio=2.0, # Further pairs ratio\n num_iters=450, # Total optimization iterations\n batch_size=8192, # Batch size for stochastic optimization\n init_method=\"pca\", # Initialization method\n optimizer=\"adam\", # Optimizer type\n lr=0.01, # Learning rate\n device=\"cuda\", # CUDA device\n low_dist_thres=1.0, # LocalMAP distance threshold\n random_state=None, # Random seed\n verbose=False # Print progress\n)\n```\n\n#### Key Methods\n\n- `fit_transform(X)`: Fit and return embedding\n- `transform(X)`: Transform new data (not implemented)\n\n#### Attributes\n\n- `embedding_`: Final embedding coordinates\n- `is_fitted`: Whether model has been fitted\n- `n_samples`: Number of training samples\n\n### Utility Functions\n\n```python\nfrom pacmap_gpu import check_gpu_availability, get_gpu_info, clear_gpu_memory\n\n# Check GPU status\ngpu_available = check_gpu_availability()\ngpu_info = get_gpu_info()\n\n# Memory management\nclear_gpu_memory()\n\n# Installation verification\nfrom pacmap_gpu import verify_installation, print_installation_report\nprint_installation_report()\n```\n\n## \u26a1 Performance Tips\n\n### Dataset Size Guidelines\n- **Small datasets (< 1K)**: GPU overhead may impact performance, but still required\n- **Medium datasets (1K-10K)**: GPU shows significant speedup over traditional CPU methods \n- **Large datasets (10K+)**: Maximum GPU performance advantage\n\n### Memory Optimization\n```python\nfrom pacmap_gpu import clear_gpu_memory\n\n# Clear memory between runs\nclear_gpu_memory()\n\n# Use efficient data types\nX = X.astype(np.float32) # Use float32 instead of float64\n```\n\n### Hardware Optimization\n- Use modern GPUs with Tensor Cores (RTX series, V100, A100)\n- Ensure sufficient GPU memory (8GB+ for datasets > 50K samples)\n- Use CUDA 12.0+ for best performance\n\n## \ud83e\uddea Running Tests and Benchmarks\n\n```bash\n# Install development dependencies\npip install -e \".[dev]\"\n\n# Run basic tests\npytest tests/ -v\n\n# Run GPU tests (requires GPU)\npytest tests/ -v -m gpu\n\n# Run benchmarks\npytest tests/test_benchmarks.py -v -m benchmark\n\n# Run without slow tests\npytest tests/ -v -m \"not slow\"\n```\n\n### Example Benchmark Output\n\n```bash\npytest tests/test_benchmarks.py::TestPerformanceBenchmarks::test_gpu_vs_cpu_performance -v -s\n\n=== MEDIUM Dataset Benchmark ===\nDataset shape: (2000, 50)\nGPU time: 1.8432s\nCPU time: 14.2156s\nSpeedup: 7.71x\n```\n\n## \ud83d\udc1b Troubleshooting\n\n### Common Issues\n\n1. **CUDA Not Found**\n ```bash\n # Check CUDA installation\n nvidia-smi\n python -c \"import torch; print(torch.cuda.is_available())\"\n ```\n\n2. **Out of Memory Error**\n ```python\n # Reduce batch size or data size\n reducer = PaCMAPGPU(batch_size=2048, n_neighbors=5)\n ```\n\n3. **Import Errors**\n ```bash\n # Verify installation\n pacmap-gpu-check\n \n # Reinstall if needed\n pip install --force-reinstall pacmap-gpu\n ```\n\n### Performance Issues\n\n- **Slow on small datasets**: GPU overhead exists, but GPU is still required for functionality\n- **GPU underutilized**: Increase dataset size or batch parameters\n- **Memory warnings**: Monitor GPU memory and clear caches regularly\n\n## \ud83e\udd1d Contributing\n\nWe welcome contributions! Please follow these steps:\n\n1. Fork the repository\n2. Create a feature branch (`git checkout -b feature/amazing-feature`)\n3. Make changes and add tests\n4. Run the test suite (`pytest tests/`)\n5. Commit changes (`git commit -m 'Add amazing feature'`)\n6. Push to branch (`git push origin feature/amazing-feature`)\n7. Open a Pull Request\n\n### Development Setup\n\n```bash\n# Clone repository\ngit clone https://github.com/AntonioRoye/PaCMAP-GPU.git\ncd PaCMAP-GPU\n\n# Set up conda environment with all dependencies (recommended)\nconda create -n pacmap-gpu-dev -c rapidsai -c conda-forge \\\n cuml-cu12 cudf-cu12 cupy-cuda12x faiss-gpu \\\n python=3.10 pytorch pytorch-cuda=12.1 -c pytorch -c nvidia\nconda activate pacmap-gpu-dev\n\n# Install package and development tools\npip install -e \".[dev]\"\n\n# Or alternatively, install from requirements file + dev extras\n# pip install -r requirements-cuda12.txt\n# pip install -e \".[dev]\"\n\n# Run tests\npytest tests/\n\n# Run GPU-specific tests (requires CUDA)\npytest tests/ -m gpu\n\n# Format code\nblack source/\n\n# Type checking\nmypy source/\n```\n\n## \ud83d\udcda Citation\n\nIf you use PaCMAP-GPU in your research, please cite:\n\n```bibtex\n@article{JMLR:v22:20-1061,\n author = {Yingfan Wang and Haiyang Huang and Cynthia Rudin and Yaron Shaposhnik},\n title = {Understanding How Dimension Reduction Tools Work: An Empirical Approach to Deciphering t-SNE, UMAP, TriMap, and PaCMAP for Data Visualization},\n journal = {Journal of Machine Learning Research},\n year = {2021},\n volume = {22},\n number = {201},\n pages = {1-73},\n url = {http://jmlr.org/papers/v22/20-1061.html}\n}\n\n@software{pacmap_gpu2025,\n title={PaCMAP-GPU: GPU-Accelerated Pairwise Controlled Manifold Approximation},\n author={Antonio Roye-Azar},\n year={2025},\n url={https://github.com/AntonioRoye/PaCMAP-GPU}\n}\n```\n\n## \ud83d\udcc4 License\n\nThis project is licensed under the Apache License 2.0 - see the [LICENSE](LICENSE) file for details.\n\n## \ud83d\udd17 Links\n\n- **Original PaCMAP**: [https://github.com/YingfanWang/PaCMAP](https://github.com/YingfanWang/PaCMAP)\n- **PyTorch**: [https://pytorch.org/](https://pytorch.org/)\n- **RAPIDS cuML**: [https://docs.rapids.ai/api/cuml/stable/](https://docs.rapids.ai/api/cuml/stable/)\n- **Issues**: [https://github.com/AntonioRoye/PaCMAP-GPU/issues](https://github.com/AntonioRoye/PaCMAP-GPU/issues)\n\n## \ud83d\ude4f Acknowledgments\n\n- Original PaCMAP authors for the algorithm and reference implementation\n- RAPIDS team for GPU acceleration libraries\n- PyTorch team for the deep learning framework\n- NVIDIA for CUDA and GPU computing platform\n\n---\n\n**Made with \u2764\ufe0f for the GPU computing community** \n",
"bugtrack_url": null,
"license": "Apache-2.0",
"summary": "GPU-Accelerated Pairwise Controlled Manifold Approximation for large-scale dimensionality reduction",
"version": "1.0.2",
"project_urls": {
"Documentation": "https://github.com/AntonioRoye/PaCMAP-GPU#readme",
"Homepage": "https://github.com/AntonioRoye/PaCMAP-GPU",
"Issues": "https://github.com/AntonioRoye/PaCMAP-GPU/issues",
"Repository": "https://github.com/AntonioRoye/PaCMAP-GPU"
},
"split_keywords": [
"dimensionality-reduction",
" manifold-learning",
" gpu",
" cuda",
" machine-learning"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "ff3a481e9237da7f27136f9c01df89f8965fca390c8f3e6bde71b09706c2e168",
"md5": "71f1fe69dbd30d2f34e80b47bdcd6702",
"sha256": "896d3e3df30eb0bdf3d81912f7c10dc2e3f5ce4381622e30c0c2df1e7a322607"
},
"downloads": -1,
"filename": "pacmap_gpu-1.0.2-py3-none-any.whl",
"has_sig": false,
"md5_digest": "71f1fe69dbd30d2f34e80b47bdcd6702",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.10",
"size": 20453,
"upload_time": "2025-07-22T03:06:03",
"upload_time_iso_8601": "2025-07-22T03:06:03.662793Z",
"url": "https://files.pythonhosted.org/packages/ff/3a/481e9237da7f27136f9c01df89f8965fca390c8f3e6bde71b09706c2e168/pacmap_gpu-1.0.2-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "2ab9c49ee9c570631eac95ec89cb362eea11460d36befe000282edf2e92e79fb",
"md5": "f615dedd033e908d1fc5e10a0545ff81",
"sha256": "d6b554786bc59788224de31fe2182721ed1104b02bd316fbeccb469ec76e97a7"
},
"downloads": -1,
"filename": "pacmap_gpu-1.0.2.tar.gz",
"has_sig": false,
"md5_digest": "f615dedd033e908d1fc5e10a0545ff81",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.10",
"size": 42076,
"upload_time": "2025-07-22T03:06:04",
"upload_time_iso_8601": "2025-07-22T03:06:04.862688Z",
"url": "https://files.pythonhosted.org/packages/2a/b9/c49ee9c570631eac95ec89cb362eea11460d36befe000282edf2e92e79fb/pacmap_gpu-1.0.2.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-07-22 03:06:04",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "AntonioRoye",
"github_project": "PaCMAP-GPU",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "pacmap-gpu"
}
JSON
