Overview
Walk-ER: A tool for converting Entity-Relationship Diagrams (ERDs) into search constraints.
- Source Code: https://github.com/hayesall/Walk-ER
Motivation
Goals are discussed more on the Walk-ER publication page.
Briefly: people frequently observe that “modes” are one of the trickiest parts in relational learning. We showed that there’s a similar representation for them in terms of Entity-Relationship Diagrams (ERDs), and that if a user labels which features are important in the digram, then there is a simple search algorithm that can convert them into a set of modes. Furthermore, modes produced this way seem to perform pretty well on benchmark problems.
Walk-ER uses a simple representation for ERDs:
Nodes: {Course=EntityNodeStyle, courseprof=RelationNodeStyle, Person=EntityNodeStyle, Proj=EntityNodeStyle, courseta=RelationNodeStyle, faculty=AttributeNodeStyle, project=RelationNodeStyle, sameperson=RelationNodeStyle}
Edges: {Person|sameperson=RelationEdge, courseprof|Course=RelationEdge, Person|courseta=RelationEdge, courseta|Course=RelationEdge, project|Proj=RelationEdge, Person|courseprof=RelationEdge, sameperson|Person=RelationEdge, Person|faculty=AttributeEdge, Person|project=RelationEdge}
Important: [courseprof, project, courseta, sameperson]
Target: faculty
RelatedEntities: {courseta=[Course, Person], courseprof=[Course, Person], project=[Person, Proj], sameperson=[Person]}
AttributeEntityMapping: {faculty=Person}
and then converts user preferences into the modes in a background file.