search

autonlab / auton-survival

Auton Survival - an open source package for Regression, Counterfactual Estimation, Evaluation and Phenotyping with Censored Time-to-Events
http://autonlab.github.io/auton-survival
MIT License
315 stars 74 forks source link

any way to visualize DSM features? Shap: Deepexplainer does not work with outputs #98

Closed mahyahemmat closed 1 year ago

mahyahemmat commented 1 year ago

Hi,

I'm trying to find a way to interpret the model outputs by visualizing feature importance in the output. Is there a way to do that?

shap.DeepExplainer usually works with Pytorch models but I've been having problem with making it work for a DSM model output of tuple type (scale, shape, logits).

chiragnagpal commented 1 year ago

Hi, thanks for using DSM ! The output of a survival model is a function of time. SHAP methods instead are designed to work with probabilistic outputs for binary outcomes.

One can use SHAP values by looking at the output of DSM at a fixed horizon of time, let's say prediction of 5 year survival.

There is a simple workaround to get SHAP to work with DSM, write an external function to encode the model and call the predict survival function at a fixed time horizon. You can then effectively treat the model as a binary classification model at that time horizon.

Hope this helps

mahyahemmat commented 1 year ago

Thanks, this worked with using shap.KernelExplainer !

chiragnagpal commented 1 year ago

awesome! would love to see how DSM performs on your target application! Keep in touch..!