MLGeometry-JAX 2.0.0

Higher dimensional computational geometry using machine learning software (JAX backend)

MLGeometry

Higher dimensional computational geometry using machine learning software

• Kahler geometry and Kahler-Einstein metrics

More to come.

Recent Changes

The backend of MLGeometry has been switched to 'JAX' from 'Tensorflow' due to the flexibility JAX provides and the current trend in the ML community. If you prefer the older version, please check the 'Using and Older Version' section below.

Installation

Note: This installs the CPU version of JAX by default. If you wish to use a GPU, it is recommended to install the appropriate version of JAX for your hardware before installing this package. Please refer to the official JAX installation guide for instructions.

You can install MLGeometry using one of the following methods:

Via PyPI

pip install MLGeometry-JAX

Directly from Github

pip install git+https://github.com/yidiq7/MLGeometry.git

Using an Older Version

If you prefer to use an older version of MLGeometry based on Tensorflow 2.16+ and Keras 3, you can check out the previous release (v1.2.1) here: Version 1.2.1 Release.

For an older version based on Tensorflow 2.12 and Keras 2, check Version 1.1.0 Release.

Follow the installation instructions provided in that release's documentation. The compatible versions of Python and CUDA can be found here.

Sample jupyter notebook

Citation

You can find our paper on arxiv or PMLR. If you find our paper or package useful in your research or project, please cite it as follows:

@InProceedings{pmlr-v145-douglas22a,
  title =    {Numerical Calabi-Yau metrics from holomorphic networks},
  author =       {Douglas, Michael and Lakshminarasimhan, Subramanian and Qi, Yidi},
  booktitle =    {Proceedings of the 2nd Mathematical and Scientific Machine Learning Conference},
  pages =    {223--252},
  year =     {2022},
  editor =   {Bruna, Joan and Hesthaven, Jan and Zdeborova, Lenka},
  volume =   {145},
  series =   {Proceedings of Machine Learning Research},
  month =    {16--19 Aug},
  publisher =    {PMLR},
  pdf =      {https://proceedings.mlr.press/v145/douglas22a/douglas22a.pdf},
  url =      {https://proceedings.mlr.press/v145/douglas22a.html},
}