百科.dev
全部条目在线工具AI 编程趋势榜开源项目技术资讯提交条目
登录
< 返回工具列表
F

falkon

> 编程语言
开源

大规模、支持多 GPU 的内核求解器

200 stars0 点赞1 次浏览
访问官网GitHub

工具介绍

大规模、支持多 GPU 的内核求解器

Falkon

Python implementation of the Falkon algorithm for large-scale, approximate kernel ridge regression.

The code is optimized for scalability to large datasets with tens of millions of points and beyond. Full kernel matrices are never computed explicitly so that you will not run out of memory on larger problems.

Preconditioned conjugate gradient optimization ensures that only few iterations are necessary to obtain good results.

The basic algorithm is a Nyström approximation to kernel ridge regression, which needs only three hyperparameters:

  1. The number of centers M - this controls the quality of the approximation: a higher number of centers will produce more accurate results at the expense of more computation time, and higher memory requirements.
  2. The penalty term, which controls the amount of regularization.
  3. The kernel function. A good default is always the Gaussian (RBF) kernel (falkon.kernels.GaussianKernel).

For more information about the algorithm and the optimized solver, you can download our paper: Kernel methods through the roof: handling billions of points efficiently

Documentation

The API is sklearn-like, so that Falkon should be easy to integrate in your existing code. Several worked-through examples are provided in the docs, and for more advanced usage, extensive documentation is available at https://falkonml.github.io/falkon/.

If you find a bug, please open a new issue on GitHub!

Installation

Dependencies:

  • Please install PyTorch first.
  • cmake and a C++ compiler are also needed for KeOps acceleration (optional but strongly recommended).

To install from source, you can run

bash
pip install --no-build-isolation git+https://github.com/FalkonML/falkon.git

We alternatively provide pre-built pip wheels for the following combinations of PyTorch and CUDA:

Linux cu118 cu121 cu124 cu126 cu128
torch 2.4.0 ✅ ✅ ✅
torch 2.5.0 ✅ ✅ ✅
torch 2.6.0 ✅ ✅ ✅
torch 2.7.0 ✅ ✅ ✅

For other combinations, and previous versions of Falkon, please check here for a list of supported wheels.

To install a wheel for a specific PyTorch + CUDA combination, you can run

bash
# e.g., torch 2.7.0 + CUDA 11.8
pip install falkon -f https://falkon.dibris.unige.it/torch-2.7.0_cu118.html

Similarly for CPU-only packages

bash
# e.g., torch 2.7.0 + cpu
pip install falkon -f https://falkon.dibris.unige.it/torch-2.7.0_cpu.html

More detailed installation instructions are available in the documentation.

Gradient-based Hyperparameter Optimization

New! We added a new module for automatic hyperparameter optimization. The falkon.hopt module contains the implementation of several objective functions which can be minimized with respect to Falkon's hyperparameters (notably the penalty, the kernel parameters and the centers themselves).

For more details check out our paper: Efficient Hyperparameter Tuning for Large Scale Kernel Ridge Regression, and the automatic hyperparameter tuning notebook.

References

If you find this library useful for your work, please cite the following publications:

@inproceedings{falkonlibrary2020,
    title = {Kernel methods through the roof: handling billions of points efficiently},
    author = {Meanti, Giacomo and Carratino, Luigi and Rosasco, Lorenzo and Rudi, Alessandro},
    year = {2020},
    booktitle = {Advances in Neural Information Processing Systems 32}
}
@inproceedings{falkonhopt2022,
    title = {Efficient Hyperparameter Tuning for Large Scale Kernel Ridge Regression},
    author = {Meanti, Giacomo and Carratino, Luigi and De Vito, Ernesto and Rosasco, Lorenzo},
    year = {2022},
    booktitle = {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics}
}

Issues· 6 开放

查看全部 Issues在 GitHub 打开

暂无开放 Issues,或尚未同步最近议题。

> 标签

Pythonaikernelkernel-methodslarge-scale-learning

暂无评论,来聊聊你的看法吧

> 工具信息

发布日期2026年8月1日
最后更新2026年9月18日
分类编程语言
定价开源

> 相关工具

T
TypeScript
JavaScript 的超集,为前端与全栈提供静态类型
P
Python
通用编程语言,广泛用于 Web、数据与 AI
G
Go
Google 推出的简洁高效系统语言