Docs

Combo blends three families of automated decision making into one configuration surface: a constraint programming solver to keep choices feasible, bandit optimization for online traffic where every request is a draw, and Bayesian optimization for offline sweeps where each trial is expensive. The constraint side comes from klause, the optimization and streaming statistics come from kumulant, and combo wires them together.

combo

Picks a configuration for each request, learns from how it scored, and never breaks the constraints you declared.

Built on

klause

Constraint solver over Boolean and integer variables. Samples feasible configurations and picks the best under a weighted objective.

kumulant

Streaming statistics and online bandits. Estimators update in constant time per observation and hold bounded memory.

0.3.3 · ★ 12

Libraries

koblas

Dense and sparse linear algebra. A BLAS and LAPACK subset in portable Kotlin, with the host library as an accelerator rather than a dependency.

skema

Typed schemas that double as a wire format, so one declaration serves Kotlin call sites and YAML documents alike.

kencode

Packs typed data classes into short URL-safe strings, for state that has to ride inside a character-limited surface.

1.3.0 · ★ 42

kpermute

Reversible integer permutations, for shuffling lists too big to hold in memory and for obfuscating sequential IDs.

1.2.0 · ★ 20