Insight on fixing "hessian or gradient contains nan or inf value(s). Convergence halted."

[xiangtgao]

I am running TimeVaryingCox on a dataset. When I fit the model, it gives convergence errors with the message: hessian or gradient contains nan or inf value(s). Any idea on where went wrong?

[CamDavidsonPilon]

Hi there, it could be a lot of things. It's possible at a specific time step (or for the full dataset), there is high colinearity between two or more columns. Have you tried adding a small l2 penalizer?