Mondrian Forests: Efficient Online Random Forests

Abstract

Ensembles of randomized decision trees, usually referred to as random forests, are widely used for classification and regression tasks in machine learning and statistics. Random forests achieve competitive predictive performance and are computationally efficient to train and test, making them excellent candidates for real-world prediction tasks. The most popular random forest variants (such as Breiman's random forest and extremely randomized trees) operate on batches of training data. Online methods are now in greater demand. Existing online random forests, however, require more training data than their batch counterpart to achieve comparable predictive performance. In this work, we use Mondrian processes (Roy and Teh, 2009) to construct ensembles of random decision trees we call Mondrian forests. Mondrian forests can be grown in an incremental/online fashion and remarkably, the distribution of online Mondrian forests is the same as that of batch Mondrian forests. Mondrian forests achieve competitive predictive performance comparable with existing online random forests and periodically re-trained batch random forests, while being more than an order of magnitude faster, thus representing a better computation vs accuracy tradeoff.

Cited by in corpus (13)

• Active learning for data streams: a survey
• Scaling Memory-Augmented Neural Networks with Sparse Reads and Writes
• A Machine Learning Approach to Online Fault Classification in HPC Systems
• The Case for Retraining of ML Models for IoT Device Identification at the Edge
• Kernel-Based Just-In-Time Learning for Passing Expectation Propagation Messages
• Binary Space Partitioning Forests
• On the asymptotics of random forests
• Particle Gibbs for Bayesian Additive Regression Trees
• BEDS-Bench: Behavior of EHR-models under Distributional Shift--A Benchmark
• Online Binary Space Partitioning Forests
• Interpretable Anomaly Detection with Mondrian P{ó}lya Forests on Data Streams
• Client-Based Intelligence for Resource Efficient Vehicular Big Data Transfer in Future 6G Network
• Selective Harvesting over Networks