Validation Tools Within Genie

[jeffs]

I've looked through the docs and other posts, but have found very little regarding validation.

I have captured multiple decision datasets from a few subject matter experts. In my application, the output is a single decision variable
from 0.0 (no) to 1.0 (yes).

I can load a data file, Learn the network Learn the parameters, but I don't see a way to then run the network on a data collection (the validation set) and capture the output.

I must be missing something obvious. I know that I can manually set the "evidence" values of each node, but that would take a long time to verify the test cases.

Please let me know if there is some simply way to load a data set with an existing network and output the decision.

[shooltz[BayesFusion]]

Please let me know if there is some simply way to load a data set with an existing network and output the decision.

There's no way of automating it in GeNIe. However, this can be implemented easily with SMILE, jSMILE or Smile.NET.

[jeffs]

So, do people just use Genie to build the network structure? It seems that the case manager and data loader are intended to allow for validation, but without an automated way to do this it seems that there is something missing . I figured that the squence would be quite straight forward:

a. load data
b. build/learn network structure
c. train network parameters
d. load validation data
c. run validation
f. write validation data to file
g. examine node values (predicted) vs. validation actuals

How do people validate their networks?

Do they all write a c++ application to load and test the data against a network created in Genie by hand?

[jga]

Hello Jeffs:

Have you found the way to extract a set of BN results?

I also have a data base with more than 1300 cases, with more than 20 evidences each one, and I am trying to evaluate my BN.

Thanks,

jga