wlearn-rf Python

Python estimator wrapper for the wlearn C11 random forest core.

Install

pip install wlearn-rf

Example

from wlearn_rf import RFModel

model = RFModel({"n_estimators": 100, "seed": 42})
model.fit(X, y)
pred = model.predict(X_test)
proba = model.predict_proba(X_test)

model.save("rf.wlrn")
restored = RFModel.load("rf.wlrn")

API

• RFModel(params=None).
• fit(X, y) trains a classifier by default; set task: "regression" for regression.
• predict(X), predict_proba(X), score(X, y).
• predict_quantile(X, quantile) and predict_interval(X, alpha) for regression forests.
• feature_importances(), permutation_importance(...), proximity(X), oob_score() inspect fitted forests.
• save(path=None) returns WLRN bytes and writes them when given a str or Path.
• RFModel.load(bytes_or_path) accepts WLRN bytes, str, or Path.
• get_params() / set_params(...) support estimator cloning/search.
• default_search_space(task=None) returns the AutoML search-space IR.
• dispose() releases native memory early in long-running processes.

Properties: n_features, n_classes, n_trees, is_fitted, capabilities.

Common params: n_estimators, max_depth, min_samples_split, min_samples_leaf, max_features, bootstrap, compute_oob, extra_trees, seed, task, criterion, sample_weight, class_weight, sample_rate, histogram_binning, max_bins, monotonic_cst.

save() returns a WLRN bundle. RF01 native bytes are an internal artifact inside the bundle, matching JavaScript @wlearn/rf.

Development

The canonical native source is repository root src/; py/csrc/ is generated for Python builds. Use WLEARN_PYTHON=/path/to/python make test-py from the repository root to force a specific Python environment.

Classification accepts int32 labels, including negative and nonconsecutive values. The classes property gives the probability-column order. Artifacts with a nonidentity label mapping use wlearn.rf.classifier@2; ordinal classifiers and regressors retain their existing @1 format. Both formats load in this release.