kappa.m

来自「模糊神经网络与软件计算的各章代码」· M 代码 · 共 29 行

function new_kappa = adjkappa(kappa, rmse)
% ADJKAPPA Adjust learning rate eta in SD according to history of RMSE.

inc_rate = 1.1;
dec_rate = 0.9;
leng = length(rmse);
if leng < 5,
	new_kappa = kappa;
	return;
end

diff_rmse = diff(rmse(leng-4:leng));

% Four consecute downs
if all(diff_rmse<0)
	new_kappa = kappa*inc_rate;
% Four consecute ups
%elseif all(diff_rmse>0)
%	new_kappa = kappa*dec_rate;
% down, up, down, up
%elseif diff_rmse(1)<0 & diff_rmse(2)>0 & diff_rmse(3)<0 & diff_rmse(4)>0,
%	new_kappa = kappa*dec_rate;
% up, down, up, down
elseif diff_rmse(1)>0 & diff_rmse(2)<0 & diff_rmse(3)>0 & diff_rmse(4)<0,
	new_kappa = kappa*dec_rate;
else
	new_kappa = kappa;
end