teaching/scientificComputing
Archived
3
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

[regression] finished exercise

Browse Source
This commit is contained in:
Jan Benda
2020-12-21 19:58:19 +01:00
parent 84a21ed91f
commit f5c3bb6139
13 changed files with 218 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
regression/code/gradientDescent.m
View File

@@ -13,7 +13,7 @@ function [p, ps, mses] = gradientDescent(x, y, func, p0, epsilon, threshold)
% mses: vector with the corresponding mean squared errors
p = p0;
gradient = ones(1, length(p0)) * 1000.0;
gradient = ones(size(p0)) * 1000.0;
ps = [];
mses = [];
while norm(gradient) > threshold
@@ -29,13 +29,12 @@ function mse = meanSquaredError(x, y, func, p)
end
function gradmse = meanSquaredGradient(x, y, func, p)
gradmse = zeros(size(p, 1), size(p, 2));
gradmse = zeros(size(p));
h = 1e-7; % stepsize for derivatives
ph = eye(length(p))*h; % ... for each parameter
mse = meanSquaredError(x, y, func, p);
for i = 1:length(p) % for each coordinate ...
pi = p;
pi(i) = pi(i) + h; % displace i-th parameter
msepi = meanSquaredError(x, y, func, pi);
msepi = meanSquaredError(x, y, func, p + ph(:,i));
gradmse(i) = (msepi - mse)/h;
end
end