RIVET

RIVET is the answer to a question multiparameter-persistence cannot answer the usual way. In one parameter a persistence module decomposes into intervals, so there is a barcode, so there is something to compute and vectorize and train on. In two parameters no such decomposition exists — that is a theorem, not a missing algorithm — so there is nothing complete to print.

RIVET’s response is to make the module interrogable instead of summarizing it. It computes the invariants that do exist and survive: the Hilbert function, the bigraded Betti numbers, and the barcodes of one-dimensional slices through the two-parameter plane. Then it precomputes enough structure that a slice can be moved interactively and its barcode redrawn immediately, which turns “which of the infinitely many slices matters” from a question you must answer in advance into one you explore.

That is a genuinely different posture from every other tool profiled here, and it explains the shape of the software. Michael Lesnick and Matthew Wright founded the project in 2013 and Wright wrote the initial code; a decade of contributors since have worked on the GUI, on parallelism, and on the presentation-minimization machinery that makes the interactivity affordable.

The consequence for a pipeline

An interactive visualizer is a hard thing to put in an automated workflow, and the honest reading is that this is the method’s nature rather than the tool’s shortcoming. Anything that hands a multiparameter module to a model has chosen a lossy summary, and RIVET’s design declines to make that choice for you.

What it does offer is rivet_console: the same computation, no Qt, driven from the command line and consumed by pyrivet. That is the seam through which the method reaches a script at all, and it is why the corpus builds the console and leaves the viewer alone.

Two things to know before packaging it

The name is taken. conda-forge already carries an unrelated project called rivet — a high-energy-physics analysis toolkit — which under strict channel priority would shadow this one entirely. rivet-recipe builds the package as rivet-console for that reason. This is the second name collision in the corpus after ripser-cpp and ripser-py, and unlike that one it is not even between two things in the same field.

There are no releases. No upstream tags, so anything packaging RIVET is pinning a commit and saying so. Research-grade infrastructure, exactly as the method note describes it.

GPL-3.0-or-later, which is redistributable copyleft with the isolate-and-notice obligations rather than a blocker.

In this corpus

rivet-recipe compiles the console from a pinned master commit, verified green on linux-64, and sits at the bottom of the three-recipe chain that pydowker-environment stages. The build needed three separate fixes and a memory-limited parallelism cap; the recipe note has them. The technique is multiparameter-persistence.