I have a question about DSL_pc. I am trying to learn a network with integer variables. The dataset has 24 variables and 5000 records. When I used DSL_pc, I didn't get any errors but when I invoked DSL_pattern::Print() after DSL_pc::Learn, it didn't show any undirected and directed edges, i.e., all zeros.
However if I ran Genie with the same dataset using PC algorithm, it produced a learned network (it's not a dag but a graph) with edges in it. By the way, i set "Discrete threshold" parameter as 3 because some variables have only 4 distinct values.
Is there any methods in DSL_pc or DSL_dataset that I can specify the possible distinct values for variables like I did in Genie? If ""Discrete threshold" is not a problem in DSL_pc, do you have any idea?
Thank you so much.
structure learning with SMILE
-
shooltz[BayesFusion]
- Site Admin
- Posts: 1457
- Joined: Mon Nov 26, 2007 5:51 pm