BayesFusion Support Forum

Reproduced & fixed. We will be releasing GeNIe tomorrow (Friday, May 8th).

This is the hard limit for the CPT representation.

You can have more than 31 parents if you use NoisyMAX nodes instead of CPT nodes, and enabled NoisyMAX decomposition in network properties.

For more information, refer to SMILE Wrappers Manual chapter "Noisy-MAX decomposition", in the PDF version ...

Error code -42 indicates an out-of-memory exception caught by SMILE. In your case, it's most likely at add_arc call used to connect P_{i+1} and T.

In SMILE, the single CPT can have 2^31 elements - with all binary nodes this would be a node with 30 parents.

We can't reproduce this issue. We were trying on Windows 10 and Windows 11, using the 'Figure 05.06.xdsl' example from files copied by GeNIe installer onto the local disk, and also using the 'Open from BayesBox' command (our repository at https://repo.bayesfusion.com contains this network).

On your ...

Thanks for the bug report. We were able to reproduce and fix the issue. New GeNIe build with the fix applied will be publicly available in the first week of May.

At this point, DSL_dataGenerator is not available through wrappers.

I'm sorry; I have misread your original question. The evidence is always passed per slice for the plate nodes in dynamic Bayesian networks. If you want to specify values for all slices, you have to write your own function.

This is correct, the temporal evidence API call takes the slice numbef and the outcome of the plate node for that slice.

In the spreadsheet below, column "A" has values copied directly from GeNIe. Column "I" has values generated using the following Sheets formula:
= Max(1,Min(13,-6+12*rand()+1+12*rand()))

When the number of histogram buckets is reduced, both show similar asymmetry:

skewed histograms.png

I think this is mostly the visualization issue. I will look into discretization, for now please note how Wolfram produces asymmetric histogram too.
wolfram-assymetric.png

You can check your expression with our online probability distribution visualizer at https://prob.bayesfusion.com/- it ...

Here's the quote from GeNIe's manual:

The final tab, Calibration, shows a very important measure of performance of a probabilistic model, notably the calibration curve. Because the output of a probabilistic model is a probability and this probability is useful in decision making, ideally we would ...

You have two options:

1. use an equation node and specify the bounds.
2. use a discrete node with numeric intervals over node outcomes.

For more information, refer to GeNIe manual. It is installed with the program; alternatively you can access HTML and PDF versions at:
https://support.bayesfusion ...

If the dataset used for parameter learning is complete, the EM uses one pass case counting. If equivalent sample size is zero (so initial network parameters have no influence on the output), and there are some CPT entries which have no records in the dataset, there will be 1.0/nodeOutcomeCount added ...

Support for continuous/hybrid dynamic Bayesian networks was added to PySMILE. It will be released as part of the next SMILE update. Private build is available for those users who need to work with equation nodes in DBNs.

I don't know whether this is a intrinsic computation constrain with GeNIe or my laptop's limitation (I7, 32G).

If you can share your model (and the evidence set, preferably as a case(s) defined in .xdsl) we'll be able to check the triangulation statistics using our internal tools.