Exception: FastRGF estimators are unavailable for usage

[caprone]

HI @HunterMcGushion ;)

then is fast implementation of RGF unavailable in Hunter??

[HunterMcGushion]

Great question!

I had actually never tried using FastRGFClassifier, or FastRGFRegressor, but I just tested a slightly modified version of HyperparameterHunter's RGF classification example, in which I change the two uses of RGFClassifier to FastRGFClassifier, and it works!

There are two things to keep in mind, though:

1. Using the FastRGF estimators is, of course, completely dependent on having FastRGF properly built (not just PyPI’s “rgf-python”). I recommend seeing their documentation if you’re unsure about this
   • When testing that FastRGF is properly installed, I also recommend using it on a small toy problem without HyperparameterHunter. This way you can make sure FastRGF is set up correctly before adding HyperparameterHunter
2. FastRGFClassifier and RGFClassifier are not completely interchangeable because they have different parameters. So in the example linked above, I commented out the following parameters in both the experiment and call to optimizer.set_experiment_guidelines: “algorithm”, “normalize”, and “loss”
   • Specifically, FastRGFClassifier doesn’t have the parameters “algorithm” or “normalize”, and it expects different values for “loss”

I hope this helps! Please let me know if you have any other questions, so we can get this sorted out! Thanks for your question, again!

TL;DR: Yes, FastRGF does work with HyperparameterHunter, but it must be properly installed, and remember that its parameters differ from those of RGF’s normal estimators

[caprone]

thanks @HunterMcGushion !
yes, my problem is that i'm on windows ;(...
thanks
sorted out!

[HunterMcGushion]

Glad to hear it! Thanks again for bringing up FastRGF!