Python和统计资料：适合混合发行吗？

Question

Sometimes, the data is not from a single distribution, but from several distributions.

For example, y = 0.4*X + 0.6*Y , y has 40% chance of coming from distribution X , and 60% chance of coming from distribution Y . A intro could be find here: http://www.r-bloggers.com/a-brief-introduction-to-mixture-distributions/ .

The problem is, given the dataset, is there any good way to fit them in python ?

I find a tutorial about R : http://www.r-bloggers.com/fitting-mixture-distributions-with-the-r-package-mixtools/ , but didn't find anything about python .

Answer 1

If your distributions are Gaussian, then scikit-learn has some good methods to fit to mixed distributions, so called Gaussian mixing models. There is a good explanation here . They also use expectation maximization, just like the R package mentioned in your link.