Algorithms

Eleven Rust-native families ship with zero external dependencies. Five more available via optional packages.

Rust-native (no dependencies)

Algorithm	Classification	Regression	Key parameter
`random_forest`	Yes	Yes	`n_estimators`
`extra_trees`	Yes	Yes	`n_estimators`
`decision_tree`	Yes	Yes	`max_depth`
`gradient_boosting`	Yes	Yes	`n_estimators`, `learning_rate`
`histgradient`	Yes	Yes	`n_estimators`, `max_bins`
`adaboost`	Yes	—	`n_estimators`
`logistic`	Yes	—	`C`
`linear`	—	Yes	—
`elastic_net`	—	Yes	`alpha`, `l1_ratio`
`naive_bayes`	Yes	—	—
`knn`	Yes	Yes	`n_neighbors`

External (optional packages)

Algorithm	Classification	Regression	Install
`xgboost`	Yes	Yes	`pip install "mlw[xgboost]"`
`lightgbm`	Yes	Yes	`pip install "mlw[lightgbm]"`
`catboost`	Yes	Yes	`pip install "mlw[catboost]"`
`svm`	Yes	Yes	Included (linear) / sklearn (nonlinear)

Auto selection

When algorithm="auto", the package selects based on task and data characteristics. The default is random_forest — a reliable baseline that works well across most problems without tuning.

Usage

# List all available algorithms
ml.algorithms()

# Classification only
ml.algorithms(task="classification")

# Use a specific one
model = ml.fit(s.train, "target", algorithm="xgboost", seed=42)

model <- ml_fit(s$train, "target", algorithm = "xgboost", seed = 42)

Engine selection

Each algorithm can run on multiple backends:

Engine	Description
`"auto"`	Uses Rust backend when available, falls back to sklearn/CRAN
`"ml"`	Rust backend (via PyO3). Zero external dependencies.
`"sklearn"`	scikit-learn backend (Python only)
`"r"`	CRAN packages (R only)

# Force Rust backend
model = ml.fit(s.train, "target", engine="ml", seed=42)

# Force sklearn
model = ml.fit(s.train, "target", engine="sklearn", seed=42)

# Force Rust backend
model <- ml_fit(s$train, "target", engine = "ml", seed = 42)

# Force CRAN packages
model <- ml_fit(s$train, "target", engine = "r", seed = 42)