1

I'm using scikit's Gaussian Mixture to obtain a 2 component mixture model for some 1-D data. Having obtained a model, I want to test how well a different data set matches this mixture model, or more precisely, with what probability is some data taken from this mixture model. I've tried using gmm.score() in order to obtain the log likelihood, then exponentiating it, but this give me a number greater than 1. What's the proper way to obtain probabilities in this case?

Hereami
  • 33
  • 2
  • 2
    The likelihood for a continuous distribution is a density, not a probability. See https://stats.stackexchange.com/questions/4220. – whuber Aug 25 '22 at 21:56
  • 1
    To obtain the "probability that a dataset is from a given model", one would need to give a meaning to the contraposite, i.e. to model that the data is not from a given model... – Xi'an Aug 26 '22 at 06:49

0 Answers0