Optimized risk scores Ustun & Rudin, KDD'17 On Monday we looked at the case for interpretable models, and in Wednesday’s edition of The Morning Paper we looked at CORELS which produces provably optimal rule lists for categorical assessments. Today we’ll be looking at RiskSLIM, which produces risk score models together with a proof of optimality. … Continue reading Optimized risk scores

Learning certifiably optimal rule lists for categorical data Angelino et al., JMLR 2018 Today we’re taking a closer look at CORELS, the Certifiably Optimal RulE ListS algorithm that we encountered in Rudin’s arguments for interpretable models earlier this week. We’ve been able to create rule lists (decision trees) for a long time, e.g. using CART, … Continue reading Learning certifiably optimal rule lists for categorical data

Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead Rudin et al., arXiv 2019 With thanks to Glyn Normington for pointing out this paper to me. It’s pretty clear from the title alone what Cynthia Rudin would like us to do! The paper is a mix of technical … Continue reading Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead