Skip to content

polar-high

A Python library for building and solving large indexed linear and mixed-integer programs.

polar-high lets you describe an indexed optimization model directly in polars. Variables and parameters are DataFrames, expressions are joins and group-by-sums on those frames, and the matrix is assembled into HiGHS without an intermediate matrix builder. The model can also be exported in MPS form to any other LP solver. The kernel is intentionally domain-free: it has no opinions about energy systems, supply chains, or any specific application.

Why this exists

If you create large indexed LPs, the bottleneck is not always the solver — it can be the time spent translating sets and parameters into a sparse matrix. polar-high keeps every step in polars: index sets are DataFrames, multiplications are joins, summations are group-bys, and the final coordinate-format → compressed-sparse-column pass is a few numpy calls. The intent is for matrix assembly to stay out of pure Python so build time scales with the LP size, not with the per-coefficient Python-object cost. Measured numbers for build time, solve time, and peak memory across N are in Compare → Benchmark.

You will probably like this engine if:

  • you already think in indexed sets (the way Pyomo, JuMP, GNU MathProg, GAMS, AMPL users do), and
  • you want a Python-only workflow with first-class DataFrame ergonomics (groupby, joins, where, lazy evaluation).

Where to go next

  • New here? Install → Quickstart.
  • Building a real model? Start in Concepts — the indexed-frame mental model is what you need before the API makes sense.
  • Already know Pyomo / linopy / JuMP framework? Jump to Compare for the design tradeoffs.
  • Need raw API surface? Reference.

Status

The engine is heavily tested but correct functioning is not guaranteed. See About.

The first downstream user is the FlexTool energy-system modelling toolkit. FlexTool's fleet of system tests (from earlier GNU MathProg to HiGHS implementation) has been used to test polar-high in real modelling use cases. In addition polar-high kernel has its own set of unit and system tests.