Almost finished gradient descent

regression/code/errorSurface.m

@@ -1,6 +1,3 @@
-clear
-close all
-
 load('lin_regression.mat');
 
 % compute mean squared error for a range of sloopes and intercepts:
@@ -16,14 +13,10 @@ end
 % plot the error surface:
 figure()
 [N,M] = meshgrid(intercepts, slopes);
-s = surface(M,N,error_surf);
+s = surface(M, N, error_surf);
 xlabel('slope', 'rotation', 7.5)
 ylabel('intercept', 'rotation', -22.5)
 zlabel('error')
 set(gca,'xtick', (-5:2.5:5))
 grid on
 view(3)
-
-set(gcf, 'paperunits', 'centimeters', 'papersize', [15, 15], ... 
-    'paperposition', [0., 0., 15, 15])
-saveas(gcf, 'error_surface', 'pdf')

regression/code/lsqGradient.m

@@ -1,7 +1,17 @@
-function gradient = lsqGradient(parameter, x, y)
-h = 1e-6;
+function gradient = lsqGradient(x, y, parameter)
+% The gradient of the least square error
+%
+% Arguments: x, the input values
+%            y, the corresponding measured output values
+%            parameter, vector containing slope and intercept 
+%                       as the 1st and 2nd element 
+%
+% Returns: the gradient as a vector with two elements
 
-partial_m = (lsqError([parameter(1)+h, parameter(2)],x,y) - lsqError(parameter,x,y))/ h;
-partial_n = (lsqError([parameter(1), parameter(2)+h],x,y) - lsqError(parameter,x,y))/ h;
+  h = 1e-6;   % stepsize for derivatives
 
-gradient = [partial_m, partial_n];
+  partial_m = (lsqError(x, y, [parameter(1)+h, parameter(2)]) - lsqError(x, y, parameter))/ h;
+partial_n = (lsqError(x, y, [parameter(1), parameter(2)+h]) - lsqError(x, y, parameter))/ h;
+
+  gradient = [partial_m, partial_n];
+end