Calculate posterior for svm about smile HOT 11 CLOSED

Do you have an estimate for this task? Thanks for this great job.

from smile.

If you strongly need it, we can start to work on it.

from smile.

It would be great, because probabilistic outputs helps to evaluate the errors. Can I help?

from smile.

Sure. You are very welcome to contribute!! Thank you!

from smile.

Any progress in this thread? Just to know...

from smile.

not yet. i am busy with other issues. are there great needs of this feature?

from smile.

Thanks! I was looking to compare the NN vs SVM results in an article.

from smile.

I am not sure if it is a good idea to compare NN and SVM by posteriori probabilities. With right choice of active functions in NN, it calculate posterior probabilities. But it is just an estimation in SVM with Platt's trick.

from smile.

I agree with you. The idea is not to benchmark each other probabilities, but to show use cases of classifiers that have probabilistic outputs (even if its only an estimation).

from smile.

Ok, it is there. I just push PlattScaling to master. To use it, please call

svm.trainPlattScaling(x, y);

after training a SVM (i.e. after call finish()). Then use

svm.predict(x, prob)

to get the posteriori probabilities.

from smile.

Wonderful! I am going to play with PlattScaling this weekend and try to share some results in an article!

from smile.