readme

clustering_code ,Clustering Through Ranking On Manifolds Version 0.2 Copyright by Markus Breit

 Clustering Through Ranking On Manifolds
 Version 0.2

 Copyright by Markus Breitenbach and Gregory Z. Grudic
 This code is for your personal and research use only.

 http://www.cs.colorado.edu/~grudic/
 http://ucsu.colorado.edu/~breitenm/

 This software is provided "as is," without warranty of any kind, express
 or implied.  In no event shall the authors be held liable
 for any direct, indirect, incidental, special or consequential damages
 arising out of the use of or inability to use this software.



LG_Cluster.m                     - Clustering of Data
Classify_New_Data.m              - Classification of new data (out-of-sample)
Set_Default_Learning_Paramters.m - Set learning parameters
batch_3_Spiral_3D.m              - A toy-data example (3 spirals)
batch_test_toys.m                - A toy-data example
batch_two_moon_toy_stuff.m
error_thing.m                    - Compute the error-rate for clustering


The model is a struct that looks like this:

               DST_TYPE: 1                 -- DST type (Euclidean / Dot-product)
                  SCALE: 1                 -- Data scaled?
                      A: [256x1 double]    -- Scaling...
                      B: [256x1 double]
          Class_Outlier: [1x4 struct]      -- Outliers for each class
                 F_orig: [800x4 double]    -- Original F matrix (columns selected using inc_clust)
                 F_norm: [800x4 double]    -- scaled F matrix (columns selected using inc_clust)
                      Y: [800x4 double]    -- Labels for the data (a 1 in a column i indicating that the case is in class i)
                 Y_orig: [800x4 double]    -- same thing, but a different underlying implementation
              mean_dist: [1x800 double]    -- mean distance of matrix D_M
         sort_mean_dist: [1x800 double]    -- sorted...
     ind_sort_mean_dist: [1x800 double]    -- the indices of the sorted matrix D_M - use for Outlier Detection
                 S_norm: [800x800 double]  -- normalized S matrix
         ind_mat_unseen: [800x800 double]  -- precomputed for unseen data
    one_over_2_sigma_sq: 0.3953            -- one over 2 sigma squared...
                      X: [800x256 double]  -- the training data
                     Dt: [1x800 double]    -- Diagonal of W
            num_classes: 4                 -- Number of clusters
           norm_iaS_inv: [800x800 double]  -- normalized F matrix
                iaS_inv: [800x800 double]  -- F matrix
               my_alpha: 0.9949            -- Alpha
               my_sigma: 1.1246            -- Sigma
              ind_clust: [4x1 double]      -- the inidices for the centers

model.Class_Outlier is a struct containing:
    val           --
    ind           -- logical indicating class membership of the case
    sort_val      --
    ind_sort      -- indices of worst outliers - use for Outlier Detection
    plot_values   --


Known bugs:
-the optimization procedure got reimplemented on the way and the results
 for USPS got slightly worse. This is currently under investigation.