Hi, I am currently studying the usage of GMMs in pymc. I wanted to get the log likelihood of a GMM model and had problems to retrieve it by how I implemented it. Lets look into this very simple example, and this will work: ### some data: rng = np.random.default_rng(123) N = 1000 W = np.array([0…

In the meantime you can get around by not setting initval in the distribution, and instead define in initvals when calling pm.sample

Understanding implementation of initial values in the usage with GMMs

Questions v5

jessegrabowski September 6, 2024, 3:34pm 5

I believe univariate_ordered is deprecated as well. You should use ordered (and upgrade to the latest version if you’re not getting a nagging message about it)

Topic		Replies	Views
Initial values not being used when sampling v5	7	2094	July 15, 2022
Bad initial energy and setting test values for ordered transform Questions	5	669	August 27, 2019
Error using Ordered() in Mixture of Uniform Distributions Questions	14	1899	February 7, 2019
`model.initial_point()` returns inf if using `pm.distributions.transforms.Ordered()` v5 bug	2	37	August 6, 2024
Problem with fitting Multivariate Mixture of Gaussians Questions	6	3561	June 4, 2018

Understanding implementation of initial values in the usage with GMMs

Related topics