I’m trying to follow this example: Multi-output Gaussian Processes: Coregionalization models using Hamadard product — PyMC example gallery especially I’m interested in predicting the outcomes within temporal gaps for some groups assuming the outcomes for other groups are known.
Unfortunately the full GP implementation from that example is very slow for my dataset even on 4 Nvidia GPUs. (1000s of time points x 5 groups). I wonder if it is possible to use HSGP approximation to speed things up.
Yes, there are hsps approximations-you can find such examples here:Gaussian Processes: HSGP Advanced Usage — PyMC example gallery. I’ll admit I’m still learning the ins and outs of pymc implementation after proposing a model mathematicall xD
Note that you’re restricted to certain types of stationary kernels, and that if youre using periodic there’s a seperate class than the one utilized here!
@Cricket4444 were you able to get this working? So far as I can tell, HSGP can’t accept cov_fun values consisting of a Hadamard product.
No, I have not. I don’t think this is possible either.