ripser.py

ripser.py puts a Python API over the Ripser engine: pass a point cloud or a distance matrix, get persistence diagrams back as arrays. That is a smaller-sounding contribution than it is. A command-line tool that writes barcodes to standard output is a subprocess call and a parser away from anything else you want to do, and this removes both.

It is a separate project from ripser-cpp, with its own maintainers under the scikit-tda organization — Christopher Tralie and Nathaniel Saul — and its own release series. It carries Bauer’s C++ as its computational core and attributes it explicitly. Both are MIT, so the arrangement is uncomplicated.

Beyond the binding it adds things a wrapper does not have to: greedy permutation subsampling for approximate diagrams on point clouds too large to do exactly, sparse distance-matrix support, and a scikit-learn-shaped transformer for pipeline use.

The name collision

Both this and ripser-cpp install as a conda package literally called ripser — this one from conda-forge, that one from Bioconda — and the version series are far enough apart that a pin looks like it disambiguates when it only narrows. The channel is the actual identity. Anyone writing ripser into a manifest should decide which project they mean before deciding which version.

In this corpus

ripser-py-environment resolves entirely from conda-forge and needs no recipe, but a single conda-forge package does not earn an automatic container the way a Bioconda one does, so it grades below its sibling despite being the easier thing to use. scikit-tda bundles it. The technique is persistent-homology.