search

ragulpr / wtte-rnn

WTTE-RNN a framework for churn and time to event prediction
MIT License
762 stars 186 forks source link

Stability of loss function for left censored data #66

Closed sharidd closed 4 years ago

sharidd commented 4 years ago

Thanks for the great blog post, got a detailed understanding of why/what of WTTE-RNN. I've recently started working with left censored data (observed labels are higher than the true labels for censored instances) and planned to implement a similar methodology. My initial enthusiasm was quickly killed by looking at the loss function ->

image

I still went ahead and implemented this loss function but am running into numerical instability issues with t/alpha or beta/alpha approaching values close to zero (after clipping these values at 0 to be able to compute log).

Just wondering if you thought about left censored data and have any recommendations to leverage this methodology.

sharidd commented 4 years ago

I went back a looked at other issues and realized better weight initialization solves the exploding loss problem, although the predicted betas are off. Closing this issue as I'll explore some steps recommended in https://github.com/ragulpr/wtte-rnn/issues/33