Added source Matlab code for reference

matlab/admixture/learn_simple_partition.m

@@ -0,0 +1,70 @@
+function part = learn_simple_partition(ordered_points, fii)
+% Goes through all the ways to divide the points into two or three groups.
+% Chooses the partition which obtains highest logml.
+
+npoints = length(ordered_points);
+
+% One cluster:
+val = calculatePopLogml(ordered_points,fii);
+bestValue = val;
+best_type = 'single';
+
+% Two clusters:
+for i=1:npoints-1
+    % The right endpoint of the first cluster.
+    val_1 = calculatePopLogml(ordered_points(1:i),fii);
+    val_2 = calculatePopLogml(ordered_points(i+1:end),fii);
+    total = val_1 + val_2;
+    if total>bestValue
+        bestValue = total;
+        best_type = 'double';
+        best_i = i;
+    end
+end
+
+% Three clusters:
+for i=1:npoints-2
+    for j=i+1:npoints-1
+        val_1 = calculatePopLogml(ordered_points(1:i),fii);
+        val_2 = calculatePopLogml(ordered_points(i+1:j),fii);
+        val_3 = calculatePopLogml(ordered_points(j+1:end),fii);
+        total = val_1 + val_2 + val_3;
+        if total>bestValue
+            bestValue = total;
+            best_type = 'triple';
+            best_i = i;
+            best_j = j;
+        end
+    end
+end
+
+part = zeros(npoints,1);
+
+switch best_type
+    case 'single'
+        part = ones(npoints,1);    
+    case 'double'
+        part(1:best_i) = 1;
+        part(best_i+1:end) = 2;
+    case 'triple'
+        part(1:best_i) = 1;
+        part(best_i+1:best_j) = 2;
+        part(best_j+1:end) = 3;
+end
+
+
+%------------------------------------------
+
+
+function val = calculatePopLogml(points,fii)
+% Calculates fuzzy (log) marginal likelihood for a population of real
+% values using estimate "fii" for the dispersion value, and Jeffreys prior
+% for the mean parameter.
+
+n = length(points);
+fuzzy_ones = sum(points);
+fuzzy_zeros = n-fuzzy_ones;
+
+val = gammaln(1) - gammaln(1 + n/fii) ...
+    + gammaln(0.5 + fuzzy_ones/fii) + gammaln(0.5 + fuzzy_zeros/fii) ...
+    - gammaln(0.5) - gammaln(0.5);