I am pretty new to Gaussian Mixture Modeling. I am following a few tutorials and all of them take a lot of time to fit. The first one is the most intuitive one import pymc as pm import numpy as np import matplotlib.pyplot as plt ## Generate data with K components. # Means, standard deviations, pr…

Neither of those code snippets run, so I can’t give you a reference for performance. Do you have a reproducible example?

Thank you for your time. You are right, I have one typo in the first snippet (y vs data in the likelihood) and one PyMC version definition problem in the second (pm.distributions.transforms.ordered vs pm.distributions.transforms.univariate_ordered). I already edited the original post. Now both snip…

So that first example takes 2 minutes to sample on my machine and it doesn’t sample particularly well given how simple the model is. Not sure that’s useful, but maybe it provides some reference. One thing that is going to slow sampling quite a bit (and hurt sampling over all) is the use of the cat…

I’d also make sure you followed the official PyMC installation guide and that you’re not using the (default and very slow) numpy BLAS. You’ll be getting a warning when you import PyMC if this is the case.

Thank you for the feedback. I checked and it is using NUTS

@jessegrabowski was asking about packages (BLAS) that live deep in the bowels of PyMC/PyTensor. The easiest way to tell is to check whether you get any warnings when you first import pymc.

You were rigth, when PyMC is imported there is this warning, WARNING (pytensor.configdefaults): g++ not detected! PyTensor will be unable to compile C-implementations and will default to Python. Performance may be severely degraded. To remove this warning, set PyTensor flags cxx to an empty string…

Are you positive you have the correct environment activated when you import pymc? Especially if you’re in an IDE like spyder. I’m asking because it shouldn’t be possible to get these errors if you install via conda-forge. If you’re sure everything is right I’d nuke the env and start over

Gausian Mixture Model takes too long to sample

Questions v5

verderis November 1, 2023, 7:00pm 10

No, Ubuntu 22.04.3 LTS

Topic		Replies	Views
Version dependant slowing down of Gaussian Mixture sampling in Ubuntu 20.04	44	878	November 7, 2023
Slow sampling speed with newer versions of PyMC v5 bug	39	1781	May 15, 2024
Importing pymc and sampling are slow on MacBook (I get blas warning) v5 installation	11	1272	August 4, 2023
Bayesian VAR example notebook: extremely low sampling rate	18	1534	May 23, 2023
Mixed multivariate Gauss distribution Questions	39	5068	April 12, 2018

Gausian Mixture Model takes too long to sample

Related topics