KeplerMapper

KeplerMapper implements Mapper, the other topological summary. Choose a lens — a function from the data to a low-dimensional space, often a projection, a density estimate, or the first components of an embedding. Cover the image with overlapping intervals. Cluster the points falling in each cover element with any scikit-learn clusterer. Build a graph whose nodes are those clusters and whose edges join clusters that share a point. The result is a network you can lay out, colour by a covariate, and look at.

That last clause is the point of the tool and the reason it sits apart from everything else profiled here. A barcode is a summary you feed to something; a Mapper graph is a summary you show a biologist. KeplerMapper leans into it — its primary output is a self-contained interactive HTML page, and the API is arranged around producing one.

What it does not inherit

Mapper is not stable in the sense persistent-homology means. The output depends on the lens, the number of cover intervals, the overlap fraction, and the clustering algorithm and its parameters, and a modest change to any of them can change the graph’s connectivity — not just its layout. There is no theorem here bounding the change in output by the change in input.

That does not make it unusable; it makes it a different kind of instrument. Mapper is for generating hypotheses you then check by another route, and a Mapper graph presented as a result without its parameter settings is not reproducible in any useful sense. The corpus keeps it as a distinct method tag rather than filing it under persistence for exactly this reason.

The upstream project classifies itself as alpha and has for a long time. Read that as a statement about the API’s stability rather than about the algorithm, which is old and well understood.

Three names for one thing

The repository is kepler-mapper, the import and conda package are kmapper, and the documentation lives under kepler-mapper.scikit-tda.org. The note and the fixture here use the package name, since that is what appears in a manifest.

In this corpus

kmapper-environment stages kmapper-recipe, which is the only thing holding it at L1 — the library is pure Python with unremarkable dependencies, and its absence from conda is a packaging gap rather than a technical one. scikit-tda bundles it. The technique is Mapper; the corpus otherwise reaches for topometry when the question is a graph-shaped view of high-dimensional biological data.