Learning Mixed Multinomial Logits with Provable Guarantees

A mixture of multinomial logits (MMNL) generalizes the single logit model, which is commonly used in predicting the probabilities of different outcomes.