Dionysus

Dionysus 2 — Dmitriy Morozov’s rewrite of the original — computes ordinary persistence competently, but that is not why it is in this corpus. It is here for the capabilities that the Rips-focused engines deliberately do not have.

Vineyards track how a persistence diagram moves as a filtration deforms, through adjacent transpositions and linear homotopies, rather than recomputing the diagram at each step. That turns persistence from a static summary into something you can follow along a parameter — a time course, a trajectory, a titration. Zigzag persistence allows the maps between complexes to point in either direction, which is what you need when a sequence of samples is not nested. Omni-field persistence computes over all prime fields at once, which is the honest answer to whether a feature you found is an artifact of working over ℤ/2. It also carries Hera for bottleneck and Wasserstein distances between diagrams.

None of those is a speed claim. Reach for ripser-cpp when the question is how fast a Rips barcode can be computed, and for this when the question is one the barcode alone cannot answer.

In this corpus

dionysus-environment resolves from conda-forge with no recipe. The technique is persistent-homology, whose implementation guide names this as the vineyards-and-zigzag option.