kernel density score VS score_samples python scikit

Question

I am using scikit learn and python for a few days now and more specially KernelDensity. Once the model is fitted I would like to evaluate the probability of new points. The method score() is made for this but apparently doesn't work as when I put an array as entry 1 number is the output. I use score_samples() but it is very slow.

I think that score is not working but I don't have skills to imrpove it. Please let me know if you have any idea

Answer 1

score() uses score_samples() as follows:

return np.sum(self.score_samples(X))

So, that's why you should use score_samples() in your case.

Answer 2

It's a bit hard to tell, without any code, but:

We assume your points you want to evaluate are saved within array X and you have a kernel density estimation kde, so you call:

logprobX = kde.score_samples(X)

But be careful, these are logarithmic! So you also need to do:

probX = np.exp(logprobX) 

These values fit to your (eventually calculated) histogram.

The time running these lines are depending on the length of X. On my machine, it's quite fast to calculate 7500 pts.