Replies: 1 comment 2 replies
-
@bbkjunior A super hacky way (and computationally expensive) way could be to use a Structural Equation Model (https://github.com/pgmpy/pgmpy/blob/dev/pgmpy/models/SEM.py) to represent a linear gaussian BN and then compute the implied covariance matrix over all the variables (https://github.com/pgmpy/pgmpy/blob/dev/pgmpy/estimators/SEMEstimator.py#L71). This would give us a joint Gaussian distribution over all the variables that you can use for doing inference. Let me know if something like this might be interesting, and I can try to write a code example. Alternatively, if you do not care about the network structure you can simply fit a multivariate gaussian distribution and do inference from it. |
Beta Was this translation helpful? Give feedback.
-
Hello!
I am aware that Continuous variables are not fully integrated yet, still, we can define e.g. LinearGaussianCPD. Is there any hacks/ugly workaround to calculate the inference with the BN with continuous variables even in the current state of the package?
Beta Was this translation helpful? Give feedback.
All reactions