Skip to content

Installation

mltreelib is a Python package with a Rust core — one wheel; pulls pyarrow>=14 (required runtime dependency).

From PyPI

pip install mltreelib

With uv

uv add mltreelib

From source

Building from source needs a Rust toolchain (rustup) and maturin:

git clone https://github.com/jkapila/mltreelib
cd mltreelib
uv run maturin develop --release   # builds the Rust core into your environment with release optimizations

Verify the install

uv run python -c "from mltreelib import ModelTree; print(ModelTree())"

An unfitted tree refuses to print its structure (by design); the call should raise ValueError: ModelTree is not fitted. Fit anything and print(model) shows the rpart-style tree.

Plotting (optional extra)

Charts and diagrams ride on the standard rendering wheels — install them only when you build figures:

pip install mltreelib[plotting]     # matplotlib + pygraphviz + seaborn
# or
uv sync --extra plotting

mltreelib.plots (metric charts) and ModelTree.to_dot() + pygraphviz (tree diagrams) are the two consumers. The core library never imports them.

What you get

ModelTree the tree: fit / predict, rpart-style text, SVG diagram, constraints
14 encoders OneHot, Hashing, Target, CatBoost, WoE, James-Stein, Count, Leave-One-Out, M-Estimator, Base-N, Sum, Quantile, TF-IDF, Embedding
scalers & builders VariableScaler, GroupedScaler, DatasetBuilder, IndexMapper
Metrics regression / classification / regression-by-group reports
dataset helpers list_datasets, load_dataset, summarize, generate, calendar / lag / window transforms

Numerics run in the Rust core; the data layer is the same code the benchmarks run on.