Learning parameters in a diagnostic model

The front end.
Post Reply
tstephens3956
Posts: 8
Joined: Tue Feb 19, 2013 9:43 pm

Learning parameters in a diagnostic model

Post by tstephens3956 »

I have developed a model for product development based on a diagnostic structure, e.g., all chance modules with targets, observations and auxiliary nodes. I have base definitions/probabilities entered into the property sheets but want to tighten up the model by learning parameters. I have tried to learn how to do this with a simple model included with tuturials the "AsiaDiagnosis" model. I have created a text data file and loaded this with the model. When I activate learn parameters, everything seems to line up well between nodes and states, however, Genie fails to learn parameters and I get an error message "em: network and dataset do not match". I do not understand why I am getting this message but suspect that it may be due to the difference in what I am doing and the very simple example in the Genie documentation which has only auxiliary nodes. Can anyone help me understand what I am doing wrong? Do target, observatory nodes need to be handled differently?
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Learning parameters in a diagnostic model

Post by shooltz[BayesFusion] »

The message you're seeing is emitted when:
a) data file contains a mix of discrete and continuous data columns
or
b) network has a mix of discrete and continuous models
or
c) network has deterministic or noisyAdder nodes which are not fixed
tstephens3956
Posts: 8
Joined: Tue Feb 19, 2013 9:43 pm

Re: Learning parameters in a diagnostic model

Post by tstephens3956 »

You were correct with option c. There was one deterministic node. Is it true that I can set this and then solve for the other parameters? Do I need to eliminate the column of data for that node or can the node state still be set by discretized data?
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Learning parameters in a diagnostic model

Post by shooltz[BayesFusion] »

tstephens3956 wrote:You were correct with option c. There was one deterministic node. Is it true that I can set this and then solve for the other parameters? Do I need to eliminate the column of data for that node or can the node state still be set by discretized data?
You'll need to add mark the deterministic node as fixed. Click the "Fixed nodes" button in the lower-left corner of "Match Network and Data" window and proceed accordingly.
tstephens3956
Posts: 8
Joined: Tue Feb 19, 2013 9:43 pm

Re: Learning parameters in a diagnostic model

Post by tstephens3956 »

Yes, that worked. I wonder while you are on a role if I can ask another question encountered on my way to this learning example. In trying to import data from an Excel file, I get essentially only an empty table. I have worked the tutorial example importing data from Microsoft Access and that works for me resulting in a table that I can adjust for missing values etc, however, Excel does not. All I get with Excel is an empty table.
Post Reply