Continuous value evidence discretisation for DBNs

The engine.
Post Reply
Marijn Peppelman
Posts: 21
Joined: Mon Jul 08, 2019 3:14 pm

Continuous value evidence discretisation for DBNs

Post by Marijn Peppelman » Wed Jan 08, 2020 3:31 pm

Hello.

First of all, happy new year.
Second, the following:

I'm aware that DBNs don't support continuous/equation nodes, and as far as i know this is not planned for the near future.

I was wondering if it is possible to use the same discretization functionality that equation nodes have to translate continuous value evidence to discrete states for chance nodes.

I'm doing a project that includes automated DBN structure optimization by way of a genetic algorithm, and i was thinking of implementing the discretization in the DBN and then also optimizing the discretization intervals along with the structure.
This would allow me to still provide continuous value evidence to the DBN and avoid having to generate new pre-discretized data sets with different discretization intervals for every potential structure.

shooltz[BayesFusion]
Site Admin
Posts: 1241
Joined: Mon Nov 26, 2007 5:51 pm

Re: Continuous value evidence discretisation for DBNs

Post by shooltz[BayesFusion] » Thu Jan 09, 2020 2:12 pm

I don't think it's possible - equation nodes carry the discretization intervals information in their definitions, the discrete nodes do not.

In 2020 we plan to work on extending discrete nodes to include numeric thresholds instead of outcome names. When this feature is complete, you'll be able to use a continuous number to set the evidence in the discrete node.

Marijn Peppelman
Posts: 21
Joined: Mon Jul 08, 2019 3:14 pm

Re: Continuous value evidence discretisation for DBNs

Post by Marijn Peppelman » Fri Jan 10, 2020 4:34 pm

Thanks for the reply.

Given my deadlines, i will go with the "pre-discretize data" workaround then.

Post Reply