To create new operator functions

[hsph-kr]

I need to add some new operator functions. Is there a way to do it? Also the exponential function is not working and it stops the kernel.

[MaxHalford]

Hey,

Alas you can't add operators yourself. New operators have to be implemented in the core XGP library. What operators did you have in mind? Yeah the exponential function probably produces plus infinity values.

[hsph-kr]

Max, Thanks for your response. I was looking for inverse and modulus function. One more clarification, the XGPClassifier uses sigmoid func to filter out probabilities and it seems the threshold value is ~0.6. Please correct me if I am wrong. I need this in a project. thanks!

[MaxHalford]

Hey,

I'll implement these functions soon!

What do you mean by the threshold value is ~0.6? Do you mean it's not predicting anything higher? If that's what you mean then it's probably just bad/good luck! I'll implement a method so that you can get the raw output and then apply the sigmoid transform yourself.

[hsph-kr]

The classifier predicts 1 when the predict_proba values are >0.6. Hence the threshold value 0.6. I think the model.program_str_ expression when evaluated and transformed using sigmoid func gives predict_proba values. Please correct me if I am wrong.

[MaxHalford]

Well if you look here the threshold is at 0.5. The program_str_, when evaluated, gives the raw values so don't worry.

[hsph-kr]

Oh. Ok. Thanks a lot for clarification.

[MaxHalford]

Hey, quick question: by modulus function do you mean the absolute value |x| or actually x % m?

[hsph-kr]

On Tue, 5 Jun 2018 at 23:00, Max Halford ***@***.***> wrote: Hey, quick question: by modulus function do you mean the absolute value |x| or actually x % m? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AeXnJS7S9kqn4aV814BmIaTJgvWdfX32ks5t5vFcgaJpZM4UXBZg> .

Hi, I meant absolute value of x. Thanks for taking it into consideration.

[MaxHalford]

Okay cheers, I'm doing it now.

[hsph-kr]

Hey Max,
I asked you earlier about exponential operator and tried it on standardised data transformed by standard scalar() but still the kernel is crashing. Does the backend implementation of exp has issues?

[MaxHalford]

Hey,

I don't think it has. But it's pretty likely that it overflows because the exponent is too high.