Learning Accurate Decision Trees with Bandit Feedback via Quantized Gradient Descent
This addresses the problem of efficient and accurate decision tree learning for practitioners across domains, offering a versatile solution that bridges gaps in existing methods, though it is incremental in combining known techniques.
The authors tackled the challenge of learning decision trees with discrete boundaries by proposing a unified gradient-based method that works in both offline supervised and online bandit feedback settings, achieving competitive or superior accuracy to specialized supervised methods and significantly outperforming applicable state-of-the-art in bandit settings.
Decision trees provide a rich family of highly non-linear but efficient models, due to which they continue to be the go-to family of predictive models by practitioners across domains. But learning trees is challenging due to their discrete decision boundaries. The state-of-the-art (SOTA) techniques resort to (a) learning \textit{soft} trees thereby losing logarithmic inference time; or (b) using methods tailored to specific supervised learning settings, requiring access to labeled examples and loss function. In this work, by leveraging techniques like overparameterization and straight-through estimators, we propose a unified method that enables accurate end-to-end gradient based tree training and can be deployed in a variety of settings like offline supervised learning and online learning with bandit feedback. Using extensive validation on standard benchmarks, we demonstrate that our method provides best of both worlds, i.e., it is competitive to, and in some cases more accurate than methods designed \textit{specifically} for the supervised settings; and in bandit settings, where most existing tree learning techniques are not applicable, our models are still accurate and significantly outperform the applicable SOTA methods.