LoRMIkA: Local rule-based model interpretability with k-optimal associations