Ripser.py: A Lean Persistent Homology Library for Python

repository·master·Indexed 43 days ago

A Python package for computing persistence cohomology using the fast C++ Ripser engine. It supports visualizing persistence diagrams, computing representative cocycles, and performing lowerstar filtrations on images. Features include a Scikit-learn compatible Rips transformer, approximate sparse filtrations via furthest point sampling (getGreedyPerm and getApproxSparseDM), and support for different coefficient fields (e.g., Z/2, Z/3).

Tokens: 6.9K
Snippets: 26
Records: 37
Agent score: 51%

What's inside Ripser.py

Use sparse distance matrices with Ripser.py
master
Ripser.py supports sparse distance matrices using the scipy.sparse library. This enables:
- Sparse filtrations: Using approximation algorithms to increase computational speed.
- Sublevelset filtration (Watershed algorithm): Quantifying critical points in grid data like time series and images. The library includes a helper function specifically for image data to facilitate this.
https://github.com/scikit-tda/ripser.py/blob/master/paper.md
Specify arbitrary coefficient fields
master
Unlike many TDA tools that are limited to binary coefficients, Ripser.py allows you to specify arbitrary field coefficients $\mathbb{Z} / p\mathbb{Z}$ for any prime $p$. This is useful for detecting topological 'twists' in data, such as differentiating a Moebius strip from an ordinary loop.
https://github.com/scikit-tda/ripser.py/blob/master/paper.md
Retrieve representative cocycles
master
Ripser.py can return representative cocycles associated with different homology classes. This feature works for both sparse and dense filtrations over any field and can be used for topological nonlinear dimension reduction (NLDR), such as mapping a point cloud to a circle or a projective plane.
https://github.com/scikit-tda/ripser.py/blob/master/paper.md
Understand the Ripser.py interfaces
master
Ripser.py provides two distinct ways to interact with the library:
1. Lightweight and functional interface: A direct, functional approach for quick computations.
2. Object-oriented interface: Designed to follow the Scikit-Learn transformer paradigm, making it easy to integrate into machine learning pipelines.
https://github.com/scikit-tda/ripser.py/blob/master/paper.md
Compute higher order homology
master
While many TDA applications focus on 0-dimensional (connected components) and 1-dimensional (loops) homology, Ripser.py is efficient enough to compute 2-dimensional homology (voids) for modest point clouds.
https://github.com/scikit-tda/ripser.py/blob/master/paper.md
Optimize Ripser.py with robin_hood hashing
master
You can achieve up to a 30% speed-up by replacing the standard STL unordered_map with the robin_hood implementation. While the PyPI package is compiled with robin_hood by default, you can manually enable it during a local build by cloning the robin_hood repository into your project root before installing ripser.py.
1. Clone the hashing implementation into a directory named robinhood at the project root: git clone https://github.com/martinus/robin-hood-hashing robinhood
2. Install ripser.py with verbose output to verify the detection: pip install -v .
```
git clone https://github.com/martinus/robin-hood-hashing robinhood
pip install -v .
```
https://github.com/scikit-tda/ripser.py/blob/master/README.md
Compile Ripser.py from source
master
To compile the package directly from the source code, clone the repository and install it in editable mode using pip.
```
git clone https://github.com/scikit-tda/ripser.py
cd ripser.py
pip install -e .
```
https://github.com/scikit-tda/ripser.py/blob/master/docs/index.rst
Install Ripser via PyPI
master
To install Ripser.py from PyPI, you must first install Cython as a prerequisite. Then, you can install Ripser using pip.
```
pip install Cython
pip install Ripser
```
https://github.com/scikit-tda/ripser.py/blob/master/docs/index.rst
Build Ripser.py from source
master
If you need to develop the package or if the pip installation fails, you can build it locally. Clone the repository and run one of the following commands from the root directory:
- For a standard local installation: pip install .
- For an editable installation (useful for development): pip install -e .
Dependencies:
- Cython
- numpy
- scipy
- scikit-learn
- persim
Platform-specific notes:
- Windows: You may need to install MinGW.
- Mac: Ensure Xcode and Xcode command line tools are updated.
```
pip install .
# OR
pip install -e .
```
https://github.com/scikit-tda/ripser.py/blob/master/README.md
Optimize performance with robin_hood_hashmap during source compilation
master
For improved computational performance when compiling from source, you can integrate the robin_hood_hashmap library. To do this, clone the robin-hood-hashing repository into the ripser/robinhood directory before compiling.
```
git clone https://github.com/martinus/robin-hood-hashing ripser/robinhood
```
https://github.com/scikit-tda/ripser.py/blob/master/docs/index.rst
Install Ripser.py
master
Ripser.py can be installed easily via pip. It provides a Python interface to the Ripser C++ library for computing Vietoris Rips (VR) filtrations and persistent homology.
```
pip install ripser
```
https://github.com/scikit-tda/ripser.py/blob/master/paper.md
Install Ripser.py via pip
master
To install the pre-compiled wheels for your platform, use the following command:
```
pip install ripser
```
https://github.com/scikit-tda/ripser.py/blob/master/README.md