- start with a random structure and compute its score
- generate all local structures of current structure:
- structures are local by doing 1 of the following to the current structure:
- add edge
- remove edge
- reverse edge
- structures are local by doing 1 of the following to the current structure:
- for each local structure compute its score
- choose structure with highest score
- repeat until convergence
Example Diagram
---structured-probabilistic-models-(spm)/pgm---model-learning-(structure-and／or-parameter)/structure-learning/structure-learning---local-search/structure-learning-local-search.png)