oqboost 2.3.0

OQBoost 2.0 — gradient-boosted 2D-oblique decision trees (histogram-binned, C++ backend)

OQBoost 2.0

Gradient-boosted 2D-oblique decision trees — histogram-binned, C++ backend.

OQBoost splits on oblique hyperplanes over feature pairs (a·u + b·v < t) instead of axis-aligned thresholds, so diagonal and interaction boundaries are represented directly rather than as axis-aligned approximations. Version 2.0 finds split directions by H-weighted least-squares regression of the gradient — deterministic, one 2×2 solve per feature pair (no random projections or numerical search).

scikit-learn compatible · compiled C++ (pybind11) + OpenMP · native missing-value handling (NaN routed to a learned bin) · pandas / scipy-sparse inputs.

Install

pip install oqboost

Prebuilt wheels for Windows, macOS (arm64), and Linux. On other platforms pip builds from source — needs a C++17 compiler and (for parallelism) OpenMP.

Quickstart

from oqboost import OQBoostClassifier, OQBoostRegressor

# Binary / multiclass classification (3+ classes use a joint softmax automatically)
clf = OQBoostClassifier(n_estimators=120, learning_rate=0.06, max_depth=4)
clf.fit(X_train, y_train)
proba = clf.predict_proba(X_test)   # (n_samples, n_classes), rows sum to 1
pred  = clf.predict(X_test)

# Regression
reg = OQBoostRegressor().fit(X_train, y_train)
y_hat = reg.predict(X_test)

Both are drop-in scikit-learn estimators — usable in Pipeline, GridSearchCV, cross_val_score, and clone; pickle / joblib compatible.

Key parameters

OQBoostClassifier(
    n_estimators=120,      # boosting rounds
    learning_rate=0.06,    # shrinkage
    max_depth=4,           # tree depth (stacked 2D oblique cuts)
    reg_lambda=1.0,        # L2 regularization
    subsample=0.8,         # row sampling per tree
    colsample=0.8,         # feature sampling per node
    multiclass="joint",    # "joint" softmax (default) | "ovr" one-vs-rest
    fast_dir="full",       # pair search: "full" all pairs (default) | "fast" Star (cheaper at high d)
)

Common extras: class_weight="balanced" for imbalance, categorical_features=[...] (cross-fitted target encoding), monotone_constraints=[...], n_iter_no_change=10 for early stopping, warm_start=True to add trees incrementally. The regressor takes the same core knobs plus loss="squared"|"huber"|"quantile".

Tips: keep max_bins small (default 16). On high-dimensional data the default fast_dir="full" is O(d²) per node — switch to fast_dir="fast" or set n_screen (feature screening) to cut training time.

Documentation

Full documentation is on GitHub: https://github.com/cree1116/oqboost-2.0

• Quickstart · Benchmarks · Explainability
• API — Classifier · Regressor · Plotting
• Guides — Categorical · Monotonic · Early stopping · Warm start · Multiclass
• Internals — Algorithm · LOB · Roadmap

OQBoost 2.0 is a ground-up rewrite. The original 1.x line (oblique splits via a Deterministic Gradient-Covariance Scan) lives at cree1116/OQBoost.

---

MIT License.