pyDowker

The filtrations elsewhere in this corpus start from a square matrix — a distance, or a directed adjacency — which is to say from things of one kind compared to each other. A great deal of biological data is not that. Cells by genes, ligands by receptors, samples by taxa, residues across an interface: relations between two different kinds of thing, whose honest description is a rectangular incidence matrix.

The Dowker complex filters exactly that. Given a relation between two sets, it builds a simplicial complex on one side whose simplices record shared relationships on the other; the classical result is that doing it from either side gives the same homology, so the asymmetry of the input does not produce an arbitrary choice in the output. pyDowker builds these from a numpy array with a user-supplied relation, filters by sublevel sets of the entries, and computes persistence — through gudhi in one parameter and through rivet in two.

The two-parameter route is the reason it sits in this family. A relation with two functions on it — an interaction strength and a confidence, say — is a bifiltration without contrivance, which is a better motivation for multiparameter persistence than the usual scale-and-density example.

Read its maturity honestly

pyDowker is a research artefact accompanying work by Niklas Hellmer and Jan Spaliński. It is a handful of commits, it says “under development” in its own README, and it has almost no users. That is not a criticism — it is what a new construction looks like before anyone has hardened it — but it sets the expectation correctly: this is a capability to evaluate, not a dependency to build a service on.

Its packaging shows the same stage. The project declares no build backend and no runtime dependencies at all, so pydowker-recipe supplies both. And __init__.py is empty, which is what let an import-only check pass while the real module — pyDowker.DowkerComplex, which reaches for pyrivet and the engine below it — could not load.

In this corpus

The leaf of the three-recipe chain pydowker-environment stages, and the fixture is the corpus’s only route into multiparameter-persistence. The recipe directory is lowercase pydowker because conda package names are; the import keeps upstream’s pyDowker, which was never ours to choose.