Dear staff,
I just find the parameter learning of GeNIe 4.0 is a bit different from GeNIe 3.0. Specifically, in 3.0 version, after I select "Randomize" for
parameter initialization, the value of Log(p) will change every time I run the model. Normally, I will run the model several times until the value of Log(p) becomes stable. However, in 4.0 version, this will not change anymore. This will lead to the different results of the BN created by 3.0 version vs 4.0 version. Especially for the model validation, I find based on the same structure of BN and same data for validation, the values in confusion matrix are significant different in version 3.0 vs 4.0. Could you please explain this situation? It makes me not confident about the model.
Thanks.
Kind regards,
Yan
parameter learning and validation in GeNIe 4.0 vs 3.0
-
- Site Admin
- Posts: 1419
- Joined: Mon Nov 26, 2007 5:51 pm
Re: parameter learning and validation in GeNIe 4.0 vs 3.0
Is there any missing data items in the learning data set? In GeNIe 4.0 when the data is complete, the EM procedure switches to simple counting, regardless of the parameter initialization.
Re: parameter learning and validation in GeNIe 4.0 vs 3.0
There's no missing data. So in this case, which one is reliable? Or I can follow the results of either version? Thanks.shooltz[BayesFusion] wrote: ↑Thu Jun 08, 2023 10:06 pm Is there any missing data items in the learning data set? In GeNIe 4.0 when the data is complete, the EM procedure switches to simple counting, regardless of the parameter initialization.
-
- Site Admin
- Posts: 431
- Joined: Tue Dec 11, 2007 4:24 pm
Re: parameter learning and validation in GeNIe 4.0 vs 3.0
I believe that case counting should be at least as good as the EM algorithm. It is a special case of EM.
I hope this helps,
Marek
I hope this helps,
Marek