I compared the performance of different inference algorithms.
The response time of the method "updateBelief" is timed after setting one evidence in a bayesnet.
In principle, evidence is set top down in my bayesnet.
The basic structure of a bayesnet is separeted in three levels. The bayesnet contains edges between nodes of these three levels. There are no edges between nodes in the same level.
For my application I use jSMILE.
In my analyses I found out that EPIS Sampling and AIS Sampling have a very bad performance in my bayesnet.
Although in a couple of papers these algorithms are described as one of the most performant sampling algorithms.
And now my question:
Do anybody know the reasons why EPIS Sampling and AIS Sampling have a bad performance in comparison to the other sampling algorithms?
Thanks in advance,
1 post • Page 1 of 1