分类器设计之logistic回归分析Ｍatlab代码

% % Liscense Notification 2013 % % Redistributions in source and binary forms, with or without % modification, are totally FREE provided that you keep this keep this % notification % % Descritption: This package demostrates how to use the logistic % regression for classfication of 2D data. % Author: Ranch Lai (ranchlai@hotmail.com) % Release date: Aug 17, 2013 %/ function [w,b,cost] = logistic_train(x,y,tol,max_iter) fprintf('training started...\n'); n = size(x,1); N = size(x,2); w = ones(n,1)/n; b = 1; cost = []; count = 1; while 1 %find gradient partial_w = zeros(n,1); partial_b = 0; for i=1:N a = exp(w'*x(:,i)+b); partial_w = partial_w + (1/(a+1)-y(i))*(-1)/((1+a)*(1+a))*a*x(:,i); partial_b = partial_b + (1/(a+1)-y(i))*(-1)/((1+a)*(1+a))*a; end old_cost = logistic_cost(x,y,w,b); step = 1; while step > 1e-12 w1 = w - step*partial_w; b1 = b - step*partial_b; new_cost = logistic_cost(x,y,w1,b1); if new_cost < old_cost break; end step = step * 0.1; end if step <= 1e-12 fprintf('finished seraching the step size\n'); break; end w = w1; b = b1; cost = [cost,new_cost]; if new_cost < tol fprintf('converged after %d iterates!',count); break; end if count > max_iter fprintf('training stoped after %d iterates, not converged to desired precision!',count); break; end count = count + 1; end