Recently found Egil Martinsson's WTTE-RNN models (https://ragulpr.github.io/2016/12/22/WTTE-RNN-Hackless-churn...), and tried to simplify all of his great work into a bare-bones example implemented in Keras. Hope somebody finds this helpful. It appears to be a very promising model for making predictions in a time-to-event (engineering failures, customer churn) context.