crospint: Coordinate Rotation for Spatial Interpolation

This package provides lightweight utilities and scikit-learn-compatible transformers to build efficient machine learning pipelines for spatio-temporal interpolation with tree-based ensemble methods. This package also contains functions and classes uses to model housing prices at the national scale (see Meslin [2026]).

This package has three distinctive features:

• It offers a native sklearn.Pipeline integration;
• It uses polars for fast feature engineering but is also fully pandas-compatible to simplify use with other ML libraries;
• It is agnostic to the algorithms used for interpolation (though is uses lightgbm by default).

This package also contains functions and classes uses to model housing prices at the national scale (see Meslin [2026]).

Installation

Just execute uv add crospint or pip install crospint.

Basic use

To be completed

Notebooks

Two notebooks illustrate the valuation method describe in Meslin [2026]:

• How to apply the conditional tail removal procedure

• How to train price models at the national scale