predict output = 'response'

[germayneng]

Hi,

Thank you for ths new methodology. Was trying out and i realized that after modelling in glm() mode, there is no way i can output my prediction in probability. I tried setting: predict(xtest, type = "response")

[matloff]

Working on this, should add this enhancement this week.

[robinyancey]

Hi @germayneng, "type=response"is not a possible input to polyfit.predict because uses the "response" type for probability internally and outputs the class with highest probability. Do you still need more information to be output?

[germayneng]

i was thinking of using it for a classification problem and hence was thinking if there is a method i could make it work with binary output :)

[robinyancey]

Ok, so does the current output fit your needs or would you prefer something else?

[matloff]

It is not clear what kind of probability output you'd like to have. With just 2 classes, there would be the conditional probability of Class 1, say, but the situation changes with multiple classes. Perhaps you'd like the set of k conditional probabilities, for k classes; under One Vs. All, we return argmax of that vector; maybe you'd like the vector itself?

[germayneng]

sorry, i think what i mean is a logit output or output like logistic regression [0,1]

[matloff]

Right, but what would you like for k > 2?

[germayneng]

i think it would be great for people who wants for multi classification :)