Learning from the existing data

[schaidee]

Hi,

I have just tried to use Spearmint for my optimization problem with a large number of parameters. In my problem, I have a black-box function which should be a noised-free function. From the characteristics of this function, I can know that which point gives the small function value using the method of my studies (but we don't know the lowest value). It means that I can choose some initial points, and compute the function value of those points.

Using Spearmint, can I set the parameters to start the optimization using the existing data (considered as prior information) instead of starting from arbitrary random points provided in the Spearmint?

Thank you very much for the answer in advance.

[JasperSnoek]

It's always a good idea to reuse existing experiments if you can!

If you use the 'spearmint lite' code, you can add as many rows into the flat results file as you like. It will automatically start using these after you added them in.