We have two slightly different versions of alignment approaches. The input to represent the weight matrix for each domain can be given in two forms:
- a sparse matrix.
- an array modeling the k-nearest neighbor for each instance. The files for this are in Alignment/knn/.
- AffineMatching.m
- Procrustes.m
- CCATwo.m
- CCAThree.m
- wmapGeneralThree.m
- wmapGeneralTwo.m
- wmapGeneralThreeInstance.m
- wmapGeneralTwoInstance.m
The code is the same as feature-level manifold projections. The only difference is how to create the correspondence matrix.
- generateWeight3.m, used to generate weight matrix, calls:
- computeOptimalMatch.m
- decompose3.m
- cmpEmbedding.m - compares different embedding results. Examples are represented as columns.
- knnsearch.m - k-nearest neighbor search
- LaplacianEigenmaps.m
- LPP.m - Locality Preserving Projections
- Showtopics.m
- createAllConnectedGraph.m
- createKnnGraph.m
- L2_distance.m - All pairs Euclidean distance