Ucha

Correctness.m

@@ -0,0 +1,21 @@
+function count = Correctness(input)
+    red = evalin('base','red');
+    blue = evalin('base','blue');
+    val_r = ones(1,size(red,2));
+    val_b = -ones(1,size(blue,2));    
+
+
+    trans = evalin('base','trans');
+    b = evalin('base','b');
+    gamma = input(1);
+    if (size(input,2) == 1)
+        a = evalin('base','a');
+    else
+        a = input(2);
+    end
+
+    prediction = SVMLS([red blue],[val_r val_b], [red blue], trans, gamma, a, b);
+    result = [val_r val_b] + prediction';
+    count = nnz(~result);
+end
+

Optimizer.m

@@ -0,0 +1,35 @@
+function Optimizer()
+    trans = evalin('base','trans');
+
+    gammas = zeros(1,5);
+    values = zeros(1,5);
+    for i = 1:5
+        [gammas(i), values(i)] = fminbnd(@Correctness,2*(i-1),2*i);
+    end
+    [Y, I] = min(values);
+    gamma = gammas(I);
+    assignin('base', 'gamma', gamma);
+    return;
+
+
+    %{
+    if (trans == 1)
+        gammas = zeros(1,5);
+        values = zeros(1,5);
+        for i = 1:5
+            [gammas(i), values(i)] = fminbnd(@Correctness,2*(i-1),2*i);
+        end
+        [Y, I] = min(values);
+        gamma = gammas(I);
+        assignin('base', 'gamma', gamma);
+        return;
+    elseif (trans == 2 || trans == 3)
+        output = fminsearch(@Correctness, [1, 1]);
+        gamma = output(1);
+        a = output(2);
+        assignin('base', 'gamma', gamma);
+        assignin('base', 'a', a);
+        return;
+    end 
+    %}
+end

README.md

@@ -1,2 +1,20 @@
-# support_vector_machine
-301 Project
+# support_vector_machine
+301 Project
+
+By Andrii Luchko & Ucha Samadashvili
+
+
+Usage:
+
+-Run app.mlapp in MATLAB. Make sure all files are in the present working directory
+
+-User can generate new data by pressing �Generate New Data� which will generate new training points randomly.
+
+-While app is running user can input new training data by assigning array of points in console to �red� (class1) and �blue� (class2) where array is the form of: blue = [point1 point2 � pointN] 
+where point is the form of: 
+point1 = [x_cord ; y_cord]
+
+-User can also input new data to be classified by assigning array of points in the same format as above to �data�
+
+-User can change parameters �gamma�, �a�, �b� manually for all transformations by sliding drag bars and then clicking �Plot�. For linear transformation user can simply click �Optimize� and then �Plot� button and the program will find the best parameter.
+

SVMLS.m

@@ -1,5 +1,6 @@
 function output = SVMLS(x_data, y_data, unknown, transformation, gamma, a, b, varargin)
-    if not((size(x_data, 2) == size(y_data, 2)))
+    x_size = size(x_data, 2);
+    if not((x_size == size(y_data, 2)))
         return    
     end
     if nargin < 4
@@ -18,7 +19,6 @@
         b = 1;
     end
 
-
     if (transformation == 1)
         Fi = @(x, xx) dot(x, xx);
     elseif (transformation == 2)
@@ -31,32 +31,32 @@
         fprintf('error')
     end
 
-    Omega = zeros(size(x_data, 2), size(x_data, 2));
-    for i = 1:size(x_data, 2)
-        for j = 1:size(x_data, 2)
+    Omega = zeros(x_size, x_size);
+    for i = 1:x_size
+        for j = 1:x_size
             Omega(i,j) = y_data(i)*y_data(j)*Fi(x_data(:,i), x_data(:,j));
         end
     end
 
-    gamma_matrix = 1/gamma*eye(size(x_data, 2));
+    gamma_matrix = 1/gamma*eye(x_size);
     Omega = Omega + gamma_matrix;
 
     A = [[0 y_data]' [-1.*y_data; Omega]];
-    o = [0; ones(size(x_data, 2), 1)];
-
+    o = [0; ones(x_size, 1)];
     x = A\o;
+
+
 
     y = zeros(size(unknown,2), 1);
     for j = 1:size(unknown, 2)
         sum = 0;
-        for i = 1:size(x_data, 2)
-            sum = sum + x(i+1)*y_data(i)*Fi(x_data(:,i), unknown(:,j))+x(1);
+        for i = 1:x_size
+            sum = sum + sqrt((x(i+1))^2)*y_data(i)*Fi(x_data(:,i), unknown(:,j))+x(1);
         end
 
         y(j) = sign(sum);
     end
-
-
-
     output = y;
-end
+
+
+end

generate_data.m

@@ -1,27 +1,20 @@
+%function to generate new set of data
 function generate_data()
-    x1_cord = 1.5*rand(1, 20)+1;
-    y1_cord = 1.5*rand(1, 20)+1;
-    x2_cord = 1.5*rand(1, 20)+2;
-    y2_cord = 1.5*rand(1, 20)+2;
-    x_data = [[x1_cord; y1_cord] [x2_cord; y2_cord]];
-    y_data = [ones(1, 20) -ones(1,20)];
-    assignin('base', 'x_data', x_data)
-    assignin('base', 'y_data', y_data)
+    red_x = 4*rand(1, 20);
+    red_y = 4*rand(1, 20);
+    blue_x = 4*rand(1, 20)+3;
+    blue_y = 4*rand(1, 20)+3;
+    red = [red_x; red_y];
+    blue = [blue_x; blue_y];
+    data = [];
 
-    x = 1:0.1:4;
-    y = 1:0.1:4;
-
-    %%%create grid
-    unknown = zeros(2, 961);
-    for i = 1:31
-        for j = 1:31
-            unknown(1, 31*(i-1)+(j)) = x(i);
-            unknown(2, 31*(i-1)+(j)) = y(j);
-        end
-    end
-    assignin('base', 'unknown', unknown)
+    assignin('base','red',red);
+    assignin('base','blue',blue);
+    assignin('base','data',data);
+    %assignin('base','trans',3);
+    %assignin('base','gamma',1);
+    %assignin('base','a',1);
+    %assignin('base','b',1);
 
-    assignin('base', 'gammaValue', 1)
-    assignin('base', 'aValue', 1)
 
 end