[P] PerpetualBooster v1.9.0 - GBM with no hyperparameter tuning, now with built-in causal ML, drift detection, and conformal prediction

PerpetualML has released PerpetualBooster v1.9.0, a significant update to its Rust-based gradient boosting machine that eliminates traditional hyperparameter tuning. The framework now includes a comprehensive suite of causal machine learning tools directly integrated into its core—Double Machine Learning, meta-learners, uplift modeling, and instrumental variables—all leveraging the same budget-based generalization approach. This means users can estimate causal effects without the typical tuning overhead. The update also adds automated drift monitoring using the trained tree structure (requiring no ground truth labels) and conformalized quantile regression for reliable prediction intervals with marginal and conditional coverage.

The technical foundation is a pure Rust core that has nearly doubled in size, now supporting 19 different objectives from regression and classification to ranking. Benchmark results are striking: compared to LightGBM tuned with 100 Optuna trials, PerpetualBooster achieves comparable accuracy with up to a 405x wall-time speedup. In head-to-head competition against AutoGluon v1.2 (a leading AutoML tool), Perpetual won 18 out of 20 OpenML classification tasks, inferred predictions up to 5x faster, and avoided out-of-memory errors on three large datasets where AutoGluon failed. The package uniquely combines causal inference, calibration, drift detection, and multi-output support in a single GBM implementation available through Python/R bindings under Apache 2.0.