From Rashomon Theory to PRAXIS: Efficient Decision Tree Rashomon Sets
Researchers have developed PRAXIS, a new algorithm designed to efficiently approximate Rashomon sets for sparse decision trees. Rashomon sets represent multiple near-optimal models that can arise from standard machine learning pipelines, offering opportunities for robust decision-making and incorporating domain knowledge. PRAXIS significantly reduces the computational resources required to compute these sets, making them more accessible for real-world datasets. AI
IMPACT Enables scalable modeling of model diversity for real-world datasets, potentially improving robustness in decision-making.
