I'm wondering if there's a way to construct a simple Bayes Net with mixed data types using probflow. The model I'm thinking of would look something like:
A -> B
A -> C
B -> C
where:
A is a Categorical Distribution (~5 categories)
B is a Continuous Distribution (probably Exponential but can be Normal)
C is a Bernoulli which takes A and C and classifies as a 1 or a 0
I realize this looks a lot like a standard Logistic Regression model, but the difference is:
the inputs aren't independent (A -> B)
I'd like to use a Categorical Distribution for A instead of creating dummy variables
Do you think this is possible by inheriting from pf.CategoricalModel? If so, any pointers on building it?
Hi Brendan, thanks for the library - really nice.
I'm wondering if there's a way to construct a simple Bayes Net with mixed data types using
probflow. The model I'm thinking of would look something like:A->BA->CB->Cwhere:
Ais a Categorical Distribution (~5 categories)Bis a Continuous Distribution (probably Exponential but can be Normal)Cis a Bernoulli which takes A and C and classifies as a 1 or a 0I realize this looks a lot like a standard Logistic Regression model, but the difference is:
A->B)Ainstead of creating dummy variablesDo you think this is possible by inheriting from
pf.CategoricalModel? If so, any pointers on building it?Cheers,
Vahndi