Neural Networks combined with Survival Analysis solutions

Hi everyone,

I could not find a matlab implementation of neural network approach combined with survival analysis, for example to predict the famous customer churn (time to event, TTE).

After thorough investigation of different neural network solutions and examples here in answers and newsgroup I gave up and stumbled upon interesting approach in github: https://ragulpr.github.io/2016/12/22/WTTE-RNN-Hackless-churn-modeling/

For those lazy to read the previous link, basically the trick is and I quote " Instead of predicting the TTE itself the trick is to let your machine learning model output the parameters of a distribution" more precisely " All you need is your favorite step-to-step RNN-architecture (also called char-RNN) with a 2-dimensional positive output layer. I recommend using SoftPlus to output β and exponential activation to output α."

Even though detailed explanation provided inside the link I couldn't implement this approach in matlab. Closest to the LSTM-RNN/char-RNN I assume is the narxnet in matlab.

So far I have just used narxnet to predict the accurate TTE (time to failure in this case).

My question is:

- Have someone implemented this kind of solution in matlab to predict TTE?

- After reading the article, I would gladly use some help in creating this WTTE-RNN-kind-of prediction, for example to some simple narx dataset / failure dataset