Feature #2438

BayesNetwork protos

Added by V. Richter over 8 years ago. Updated over 8 years ago.

Status:ResolvedStart date:11/20/2015
Priority:NormalDue date:
Assignee:J. Moringen% Done:

100%

Category:Type Proposal
Target version:-

Description

I would like to add prototypes describing a bayesian network and a corresponding observation.

0001-Add-a-BayesNetwork-proto-description.patch Magnifier - BayesNetwork and BayesVariable (4.82 KB) V. Richter, 11/20/2015 12:12 PM

0002-Add-BayesNetworkEvidence-proto.patch Magnifier - BayesNetworkEvidence describing a network observation. (1.95 KB) V. Richter, 11/20/2015 12:12 PM

changes.diff Magnifier (6.93 KB) J. Moringen, 11/24/2015 12:14 PM

Associated revisions

Revision 77a25a90
Added by V. Richter over 8 years ago

Added Bayes{Variable,Network[Evidence]} message types

fixes #2438

  • proto/sandbox/rst/bayesnetwork/BayesVariable.proto: new file; variable
    in a Bayes network
  • proto/sandbox/rst/bayesnetwork/BayesNetwork.proto: new file;
    description of Bayes network equivalent to the XMLBIF Format
  • proto/sandbox/rst/bayesnetwork/BayesNetworkEvidence.proto: new file;
    the observed state of a BayesNetwork at a specific time or time span

Signed-off-by: Jan Moringen <>

Revision b592d469
Added by V. Richter over 8 years ago

Added Bayes{Variable,Network[Evidence]} message types

fixes #2438

  • proto/sandbox/rst/bayesnetwork/BayesVariable.proto: new file; variable
    in a Bayes network
  • proto/sandbox/rst/bayesnetwork/BayesNetwork.proto: new file;
    description of Bayes network equivalent to the XMLBIF Format
  • proto/sandbox/rst/bayesnetwork/BayesNetworkEvidence.proto: new file;
    the observed state of a BayesNetwork at a specific time or time span

Signed-off-by: Jan Moringen <>

History

#1 Updated by J. Moringen over 8 years ago

  • Status changed from New to In Progress
  • Assignee set to J. Moringen

Reviewing …

#2 Updated by J. Moringen over 8 years ago

Minor changes are attached.

Discussion-worthy:
  • Should the property fields be key-value collections (as suggested in the attached patch)?
  • Since there seems to be no relation to other types in the rst.graph package and the new types would just as well fit into rst.inference (if we actually had that), maybe a separate package rst.bayesnetwork would be better?

#3 Updated by V. Richter over 8 years ago

J. Moringen wrote:

Discussion-worthy:
  • Should the property fields be key-value collections (as suggested in the attached patch)?

The XMLBIF works with string properties of undefined format so using a key-value collection would hinder the interoperability with the existing format.

    <PROPERTY>size 12</PROPERTY>
    <PROPERTY>comment Trial number ten</PROPERTY>
    <PROPERTY>position = (154, 241)</PROPERTY>

  • Since there seems to be no relation to other types in the rst.graph package and the new types would just as well fit into rst.inference (if we actually had that), maybe a separate package rst.bayesnetwork would be better?

I used the rst.graph package because a bayes network is a graph. What about rst.graph.bayesnetwork?.

#4 Updated by J. Moringen over 8 years ago

  • % Done changed from 0 to 50

V. Richter wrote:

The XMLBIF works with string properties of undefined format so using a key-value collection would hinder the interoperability with the existing format.

OK.

I used the rst.graph package because a bayes network is a graph. What about rst.graph.bayesnetwork?.

I understand. My argument was that Bayesian networks should get their own package since they cannot simultaneously be contained in the multiple domains (graph, inference, graphical model, stochastics) they belong to.

#5 Updated by V. Richter over 8 years ago

I used the rst.graph package because a bayes network is a graph. What about rst.graph.bayesnetwork?.

I understand. My argument was that Bayesian networks should get their own package since they cannot simultaneously be contained in the multiple domains (graph, inference, graphical model, stochastics) they belong to.

That is a good point. In this case the rst.bayesnetwork package would be okay.

Could you change that? I do not know how to apply your diff to my project :-)

#6 Updated by J. Moringen over 8 years ago

V. Richter wrote:

That is a good point. In this case the rst.bayesnetwork package would be okay.

Could you change that? I do not know how to apply your diff to my project :-)

Sure.

#7 Updated by V. Richter over 8 years ago

  • Status changed from In Progress to Resolved
  • % Done changed from 50 to 100

Also available in: Atom PDF