Timezone: »
Poster
Harnessing the power of choices in decision tree learning
Guy Blanc · Jane Lange · Chirag Pabbaraju · Colin Sullivan · Li-Yang Tan · Mo Tiwari
We propose a simple generalization of standard and empirically successful decision tree learning algorithms such as ID3, C4.5, and CART. These algorithms, which have been central to machine learning for decades, are greedy in nature: they grow a decision tree by iteratively splitting on the best attribute. Our algorithm, Top-$k$, considers the $k$ best attributes as possible splits instead of just the single best attribute. We demonstrate, theoretically and empirically, the power of this simple generalization. We first prove a greediness hierarchy theorem showing that for every $k\in \mathbb{N}$, Top-$(k+1)$ can be dramatically more powerful than Top-$k$: there are data distributions for which the former achieves accuracy $1-\epsilon$, whereas the latter only achieves accuracy $\frac{1}{2}+\epsilon$. We then show, through extensive experiments, that Top-$k$ outperforms the two main approaches to decision tree learning: classic greedy algorithms and more recent ``optimal decision tree'' algorithms. On one hand, Top-$k$ consistently enjoys significant accuracy gains over greedy algorithms across a wide range of benchmarks. On the other hand, Top-$k$ is markedly more scalable than optimal decision tree algorithms and is able to handle dataset and feature set sizes that remain far beyond the reach of these algorithms. The code to reproduce our results is available at https://github.com/SullivanC19/pydl8.5-topk.
Author Information
Guy Blanc (Stanford University)
Jane Lange (MIT)
Chirag Pabbaraju (Stanford University)
Colin Sullivan (Computer Science Department, Stanford University)
I am a master's student at Stanford studying computer science (CS) on the artificial intelligence track. I received a bachelor's degree from Stanford studying CS on the theory track with a minor in mathematics. My work with the Thrun lab is centered on optimized learning of decision trees, and I enjoy a broader interest in interpretable artificial intelligence. I have competed at a high level in programming and AI contests such as USACO and BattleCode and further improved my development skills working for companies ranging in size from the rapidly growing Nimble Robotics to multinational corporations such as Amazon and LinkedIn.
Li-Yang Tan (Computer Science Department, Stanford University)
Mo Tiwari (Stanford University)
More from the Same Authors
-
2023 Poster: Provable benefits of score matching »
Chirag Pabbaraju · Dhruv Rohatgi · Anish Prasad Sevekari · Holden Lee · Ankur Moitra · Andrej Risteski -
2023 Poster: BanditPAM++: Faster $k$-medoids Clustering »
Mo Tiwari · Ryan Kang · Donghyun Lee · Sebastian Thrun · Ilan Shomorony · Martin Zhang -
2022 Poster: MABSplit: Faster Forest Training Using Multi-Armed Bandits »
Mo Tiwari · Ryan Kang · Jaeyong Lee · Chris Piech · Ilan Shomorony · Sebastian Thrun · Martin Zhang -
2021 Poster: Provably efficient, succinct, and precise explanations »
Guy Blanc · Jane Lange · Li-Yang Tan -
2021 Poster: Universal Approximation Using Well-Conditioned Normalizing Flows »
Holden Lee · Chirag Pabbaraju · Anish Prasad Sevekari · Andrej Risteski -
2020 Poster: BanditPAM: Almost Linear Time k-Medoids Clustering via Multi-Armed Bandits »
Mo Tiwari · Martin Zhang · James J Mayclin · Sebastian Thrun · Chris Piech · Ilan Shomorony -
2020 Poster: Efficient semidefinite-programming-based inference for binary and multi-class MRFs »
Chirag Pabbaraju · Po-Wei Wang · J. Zico Kolter -
2020 Spotlight: Efficient semidefinite-programming-based inference for binary and multi-class MRFs »
Chirag Pabbaraju · Po-Wei Wang · J. Zico Kolter -
2020 Poster: Universal guarantees for decision tree induction via a higher-order splitting criterion »
Guy Blanc · Neha Gupta · Jane Lange · Li-Yang Tan -
2020 Poster: Estimating decision tree learnability with polylogarithmic sample complexity »
Guy Blanc · Neha Gupta · Jane Lange · Li-Yang Tan -
2018 Poster: Multiple Instance Learning for Efficient Sequential Data Classification on Resource-constrained Devices »
Don Dennis · Chirag Pabbaraju · Harsha Vardhan Simhadri · Prateek Jain