Attacking Clustered Data with a Mixed Effects Random Forests Model in Python
summary
Shows a Python implementation of MERF as a pip-installable MERF with a scikit-learn style estimator and an EM loop that alternates training a random forest and cluster-level linear corrections. Evaluations use 500 training samples across 100 highly imbalanced clusters and report up to a 70% MSE gain versus a global random forest on known clusters, roughly zero harm on new clusters, and practical scaling to thousands of clusters while still converging.
AI-written summary · about
chapters (12)
- 0:00Mixed effects random forests intro
- 1:09Why mixed effects models for clusters
- 8:56Classical mixed effects limits
- 10:20MuRF Python package
- 17:25Synthetic data performance
- 24:00Real-world client results
- 26:31Future work and collaboration
- 28:20Q&A: Initial conditions & robustness
- 29:36Q&A: Unsupervised clustering & nesting
- 31:01Q&A: Feature selection
- 33:51Q&A: Scalability & high-cardinality
- 35:15Q&A: Spark vs CatBoost
related talks
27mUsing Random Forests in Python
21mWhy a 1990s Machine Learning Algorithm Destroys LLMs at...
36mAdventures in scikit-learn's Random Forest
44mUnderstanding Random Forests
57mFrom Bonsais to Random Forests
54mDebiasing Random Forests For Treatment Effect Estimation
27mMachine Learning at Scale
32mAn Introduction to 'metalearners', a Flexible MetaLearner Library in Python