I have been quiet for past six months so it is high time to write an update. In April I wrote about my Time-to-event Prediction Guide which explains how to apply survival analysis in practice. Guide is still in Preview as I have been crafting a custom algorithm that I named as Lifespan Predictor. Free Tier version of the algorithm is now available in AWS Marketplace, check it out!
I would like to get some feedback from you. What would be the most valuable next work item for you? Vote below or register to suggest your own idea in the comments.
-
Publish Lifespan Predictor for Snowflake
Packages container as a Snowflake native app and offer it via Snowflake marketplace as well.
-
Support SHAP for DeepSurv
Gives insight of what is driving the risks according to DeepSurv.
-
Support Cox model
Adds support for model_type='coxph'. Linear models are nice for explainability – hazard ratios could be returned per feature to understand what is driving the risks generally and what is driving the risks for a particular sample.
-
Support DeepHit
Adds support for model_type='deephit' and multiple failure modes. DeepHit model supports competing risks natively. This is nice if your data contains multiple failure modes. Predictions are not just one survival curve, but an array of curves.
-
Finalise the Time-to-event Prediction Guide
The guide is meant as an education material on survival analysis. It helps to get more value out of Lifespan Predictor too.
Leave a Reply