Fixes deprecated cluster_bingham.m and bingham_mlesac.m

matlab/bingham_cluster.m

@@ -0,0 +1,27 @@
+function [B weights] = bingham_cluster(X, min_points)
+% [B weights] = bingham_cluster(X, min_points) -- where each B(i) contains fields V, Z, F
+
+if nargin < 2
+   min_points = 20;
+end
+
+for i=1:100 %max no of clusters
+  fprintf('Bingham distrubution #: %d', i);
+  [B(i) outliers] = bingham_fit_mlesac(X);
+  % outliers is a row vector
+  weights(i) = size(X,1) - length(outliers);
+
+  fprintf('no of points left are', weights(i));
+  % weight of a particular bingham is no of points left - no of ouliers
+  if length(outliers) < min_points
+    break;
+  end
+  X_updater = zeros(length(outliers), size(X,2));
+  % initialize the outlier data for next iteration to zeros
+  for j = 1:length(outliers)
+    X_updater(j,:) = X(outliers(j),:);
+  end
+  % get them outliers
+  X = X_updater;
+  % update the outliers for the next iteration
+end

matlab/deprecated/bingham_fit_mlesac.m → matlab/bingham_fit_mlesac.m

@@ -1,71 +1,66 @@
-function [B outliers] = bingham_fit_mlesac(X)
-% [B outliers] = bingham_fit_mlesac(X) -- where B = B.{V, Z, F}
-
-d = size(X,1);
-n = size(X,2);
-
-iter = 100;
-p0 = 1 / surface_area_hypersphere(d-1);  % uniform density for outliers
-logp0 = log(p0);
-
-pmax = -inf;
-
-for i=1:iter
-
-   % pick d points at random from X
-   r = randperm(n);
-   r = r(1:d);
-   Xi = X(:,r);
-
-   % fit a Bingham to the d points
-   [V Z F] = bingham_fit(Xi);
-   %[V Z F] = bingham_fit_scatter(Xi*Xi')
-
-   % compute data log likelihood
-   logp = 0;
-   for j=1:n
-      p = bingham_pdf(X(:,j), V, Z, F);
-      if p > p0
-         logp = logp + log(p);
-      else
-         logp = logp + logp0;
-      end
-   end
-
-   if logp > pmax
-      pmax = logp;
-      B.V = V;
-      B.Z = Z;
-      B.F = F;
-
-      %fprintf('*** found new best with log likelihood %f ***\n', logp);
-
-      %figure(20);
-      %plot_bingham_3d(V,Z,F,X');
-      %figure(21);
-      %plot_bingham_3d_projections(V,Z,F);
-
-   end
-end
-
-% find inliers/outliers and fit the Bingham to all the inliers
-
-L = zeros(1,n);
-for j=1:n
-   p = bingham_pdf(X(:,j), B.V, B.Z, B.F);
-   if p > p0
-      L(j) = 1;
-   else
-      L(j) = 0;
-   end
-end
-
-inliers = find(L);
-outliers = find(~L);
-
-[B.V B.Z B.F] = bingham_fit(X(:,inliers));
-
-
-
-
-
+function [B outliers] = bingham_fit_mlesac(X)
+% [B outliers] = bingham_fit_mlesac(X) -- where B = B.{V, Z, F}
+X = X';
+d = size(X,1);
+n = size(X,2);
+
+iter = 100;
+p0 = 1 / surface_area_hypersphere(d-1);  % uniform density for outliers
+logp0 = log(p0);
+
+pmax = -inf;
+
+for i=1:iter
+   fprintf('mlesac iteration: %d', i);
+
+   % pick d points at random from X -> put them into X_1, X_2, X_3, X_4
+   r = randperm(n);
+   r = r(1:d);
+   for j=1:d
+        eval(['X_' num2str(j) '= X(:,r(j));'])
+   end   
+   X_combined = [X_1 X_2 X_3 X_4];
+   X_combined = X_combined';
+
+   % fit a Bingham to the d points
+   bing_X_combined = bingham_fit(X_combined);
+
+   % compute data log likelihood
+   logp = 0;
+   for j=1:n
+      p = bingham_pdf(X(:,j)', bing_X_combined);
+      if p > p0
+         logp = logp + log(p);
+      else
+         logp = logp + logp0;
+      end
+   end
+
+   % update the threshold
+   if logp > pmax
+      pmax = logp;
+   end
+
+end
+
+% find inliers/outliers
+L = zeros(1,n);
+for j=1:n
+   p = bingham_pdf(X(:,j)', bing_X_combined);
+   if p > p0
+      L(j) = 1;
+   else
+      L(j) = 0;
+   end
+end
+
+inliers = find(L);
+outliers = find(~L);
+fprintf('no of outliers were %d', outliers);
+
+% fit a Bingham to all the inliers
+bing_return = bingham_fit(X(:,inliers)');
+B = bing_return; 
+
+
+