# Libraries for Bayesian network inference with continuous data

Is there any good libraries that allow me to:

1. Construct a Bayesian network manually
2. Specify the conditional probabilities with any continuous PDF, not just Guassian
3. Perform inference, either exact or approximate

I looked at the following libraries so far, none of them meet the 3 requirements:

• pgmpy: only work on discrete distribution or linear Guassian distribution
• bnlearn: same as pgmpy
• gRain: only discrete distribution
• Huggin: only discrete distribution and Guassian
• deal: no support for inference
• abn: same as deal
• libpgm: only discrete distribution and Guassian
• Investigate tensorflow, edward, and pymc
– Emre
Apr 12 '17 at 3:25

You can use pymc3. I am pretty sure it works for all the 3 requirements. http://pymc-devs.github.io/pymc3/

• I don't see a way to construct Bayesian network (directed graphical model) using PyMC3, but it seems that Edward, which depends on PyMC3, has that support. Apr 14 '17 at 6:07
• Since Edward uses PyMC3, I am pretty sure you can build PGMs with PyMC3 directly. The implementation is not that hard. I have tried PGMPy but since you ask for any continuous pdf as your requirement, you need to use PyMC3. You just need to go a level deeper writing your conditional distributions as equations. The inference and analysis are as easy as it gets! Apr 15 '17 at 18:15

Not a library, but a interactive GUI based tool is "samiam" (Sensitivity Analysis Modeling Inference and More) from a research group at UCLA.

I am not sure about your "continuous PDFs" requirement, whether it's possible to define them inside the samiam GUI.

For API-access, you might call functions inside the inflib.jar file.