Role of hyperparameters in 20 second regression version

[patel-zeel]

I was a bit curious about what is going on in the background when we compute posterior in the 20-second version. Is there any hyperparameter tuning happening in the background? Or it is just computing the posterior with default hyperparameters.

To put it in another way, this version is ready to use in production or I need to use some additional steps to get a better posterior?

[wesselb]

Hey!

To directly answer your question, in the 20-second version, there is no hyperparameter tuning happening. It is just computing the posterior with the default hyperparameters (variance 1 and length scale also 1). If you'd like to do something like this in production, I would recommend first doing maximum likelihood to get a better estimate of the length scale and variance.

[patel-zeel]

Hi, Thanks for solving the query. I got too excited at first sight and deployed it for real-world application before I realized this :)

[wesselb]

No problem at all! I'm glad to hear that you got excited. :) If there is a need for a way to more easily and succinctly incorporate hyperparameter learning, then we could certainly work something out!

[patel-zeel]

Sure. Currently, I find it neat and clean but we can look into this as I explore more.

I must mention that you have an amazing set of decentralized tools working together (Stheno, mlkernels, varz, lab and what not!).