issue with setting evidence

The front end.
Post Reply
burnsmc
Posts: 4
Joined: Mon Mar 12, 2018 7:05 pm

issue with setting evidence

Post by burnsmc » Thu Apr 12, 2018 5:15 pm

I have a stationary network with a node 'Day' that has 365 states (one for each day of the year). Any day I select past 200 automatically sets a control value to a different state. If I set Day observed to any day between 1 and 200 it works as intended and all other values update accordingly. Why is this happening and how do I get it to stop? I have made the same network on a temporal plate and it gives me no issues past 200.

Thanks,
Meghan
Attachments
Safe2Fly_static_noLearning.xdsl
(426.09 KiB) Downloaded 195 times

shooltz[BayesFusion]
Site Admin
Posts: 1205
Joined: Mon Nov 26, 2007 5:51 pm

Re: issue with setting evidence

Post by shooltz[BayesFusion] » Thu Apr 12, 2018 8:17 pm

The problem is with GeNIe's menu command range reserved for the evidence states. As you just discovered, it's just 200 entries. Any outcomes above 200 fall out of the range and GeNIe can't handle the command properly. We'll increase the reserved range in the next release. In the meanwhile, the following workaround is feasible:

1. open the Case Manager pane from the View menu
2. uncheck the Show Only Evidence Nodes option in Case Manager's toolbar (the button has the grounding symbol icon)
3. you'll get the grid with columns for all nodes in your network
4. click on the Add New Case button on the tooolbar
5. select the appropriate node states in the menu
6. click the Apply Case button on the toolbar - this will use the evidence defined in the case and set it in the network.

marek [BayesFusion]
Site Admin
Posts: 271
Joined: Tue Dec 11, 2007 4:24 pm

Re: issue with setting evidence

Post by marek [BayesFusion] » Fri Apr 13, 2018 3:24 am

Meghan,

Thanks for finding this bug -- we will fix it in the next release. I just wanted to comment here that having 365 states is not a usual modeling practice. I don't know your problem and can't advise without knowing it but I feel that you are asking for problems by having a node with so many states. Please note that any child of this node will have at least 365 columns in the CPT. Two of such nodes being parents of a node will result in at least 100K columns. You will exhaust your memory very quickly. Please rethink this.
Good luck!

Marek

burnsmc
Posts: 4
Joined: Mon Mar 12, 2018 7:05 pm

Re: issue with setting evidence

Post by burnsmc » Fri Apr 13, 2018 12:27 pm

Thanks everyone! The workaround was successful. I realize that this is not the most efficient way to solve the problem, but that was actually the point of doing it this way as an example for my thesis.

Post Reply