Matlab Optimierungsproblem

WICHTIG: Der Betrieb von goMatlab.de wird privat finanziert fortgesetzt. - Mehr Infos...

Partner:

Forum

Option

[Erweitert]

• Diese Seite per Mail weiterempfehlen

Gehe zu:

Lisa89

Forum-Anfänger


	Beiträge: 33

	Anmeldedatum: 13.12.18

	Wohnort: ---

	Version: ---

Verfasst am: 24.11.2019, 16:37 Titel: Matlab Optimierungsproblem

Hallo,

ich suche jemanden, der mir bei der Programmierung eines Models aus der Volkswirtschaft helfen kann. Es handelt sich dabei, um ein Optimierierungsproblem.
Ich habe das Model schon programmiert nur es gibt einen Teil, den ich leider nicht verstehe. Ich habe auch einen Code, der für so einen ähnlichen Fall geschrieben worden ist aber leider kann ich den Code nicht nachvollziehen. Kann mir jemand dabei helfen?

LG
Lisa

Harald

Forum-Meister


	Beiträge: 24.492

	Anmeldedatum: 26.03.09

	Wohnort: Nähe München

	Version: ab 2017b

Verfasst am: 24.11.2019, 16:47 Titel:

Hallo,

poste am besten den Code und die konkrete Frage dazu, dann kann man sehen, ob man dir weiterhelfen kann.

Grüße,
Harald
_________________

1.) Ask MATLAB Documentation
2.) Search gomatlab.de, google.de or MATLAB Answers
3.) Ask Technical Support of MathWorks
4.) Go mad, your problem is unsolvable ;)

Lisa89

Themenstarter

Forum-Anfänger


	Beiträge: 33

	Anmeldedatum: 13.12.18

	Wohnort: ---

	Version: ---

Verfasst am: 26.11.2019, 12:34 Titel:

Hallo Harald,

der Code sieht wie folgt aus: Das ist das eigentliche Model.

Code:

% Version 1: Taylor-Rule depends on E[x(+1)] and E[pi(+1)]

% Q Number of quarters in data output
% seed Initializer for the pseudo random number generator
% varargin Additional input arguments that can be used to change standard parameterization
% e.g.: GenerateDynamics(100, 1, 'h',0.5)
% del_d_s can be used to analyze UNconventional monetary policy
%
% data Structure array containing all return data

%% Initialization
randn('seed',seed); % ... of pseudo random number generator
exploded = 0; % ... of control variable to check whether dynamics did explode

%% Taylor Rule
del_x = 0.5; % Taylor Parameter Outputgap
del_pi = 1.5; % Taylor Parameter Inflation
eta = 0; % Persistence ... + eta*i(q-1)
del_d_s = 0;

%% Financial Markets Tax
tau = 0; % Financial transaction tax
tax_type = 'FTT'; % FTT or FAT, FAT_progressive (bei letzterer muss unten der 'normale' Steuersatz gesetzt werden)

%% Shock 0, 1, 2: 0 --> No shock
% 1 --> Cost shock in q = 10
% 3 --> Interest Rate Shock
% 4 --> tau shock
Shock = 0;

%% interaction parameter
k = 0;
kap = 0;
c_1 = 0;
c_3 = 0.5;
c_4 = 1;
h = 0;

%% NKM Parameters
sig = 1;
%gam = 0.17166; % wird unten aus den deep parameters berechnet
bet = 0.99;
alp_x = 0.2;
alp_pi = 0.2;
alp_s = 0.2;
zeta = 0.5;
phi = 10;

sig_eps = 0.15;

alp=0.0; eps=6; thet=0.67; varphi=1;

%% ABC Parameter
a = 1;
b = 0.04;
c = 0.04;
d = 0.975;
e = 300;
f_bar = 0;
var_s = 0.01;
var_C = 0.05;
var_F = 0.01;

%% Special parameters
Turn_Noise_Off_in_q = -1;
ParTabPath = ''; % to which path should the parameterization table be saved ('' meens none)

% Weights that are used to bring the daily stock prices to a quartertly level
st_sq_weights = (ones(64,1)/64); % <-- equally weighted
%st_sq_weights = (1-0.5) * (0.5).^[63:-1:0]'; % <-- geometrically decaying weighted
%st_sq_weights = st_sq_weights./sum(st_sq_weights);

silent = false; % Bei explosivem Pfad keine Warnung anzeigen

%% Change parameterization depending on "nargin"
for i = 1:length(varargin)
if strcmpi(varargin{i},'h')
h = varargin{i+1};
end
if strcmpi(varargin{i},'kap')
kap = varargin{i+1};
end
if strcmpi(varargin{i},'c_1')
c_1 = varargin{i+1};
end
if strcmpi(varargin{i},'k')
k = varargin{i+1};
end
if strcmpi(varargin{i},'1_minus_thet')
thet = 1 - varargin{i+1};
end
if strcmpi(varargin{i},'alp_x')
alp_x = varargin{i+1};
end
if strcmpi(varargin{i},'phi')
phi = varargin{i+1};
end
if strcmpi(varargin{i},'zeta')
zeta = varargin{i+1};
end
if strcmpi(varargin{i},'alp_pi')
alp_pi = varargin{i+1};
end
if strcmpi(varargin{i},'del_d_s')
del_d_s = varargin{i+1};
end
if strcmpi(varargin{i},'a')
a = varargin{i+1};
end
if strcmpi(varargin{i},'b')
b = varargin{i+1};
end
if strcmpi(varargin{i},'Turn_Noise_Off_in_q')
Turn_Noise_Off_in_q = varargin{i+1};
end
if strcmpi(varargin{i},'ParTabPath')
ParTabPath = varargin{i+1};
end
if strcmpi(varargin{i},'del_x')
del_x = varargin{i+1};
end
if strcmpi(varargin{i},'del_pi')
del_pi = varargin{i+1};
end
if strcmpi(varargin{i},'eta')
eta = varargin{i+1};
end
if strcmpi(varargin{i},'silent')
silent = true;
end

end

gam = (1-thet)*(1-bet*thet)/thet * (1-alp)/(1-alp+alp*eps) * (sig + (varphi+alp)/(1-alp)); %0.17166; %0.35;

%% create data vectors
burnin = 20; % quarters to remove at beginning
T = (Q+burnin)*64; % Total number of day (including burnin)

% Data Vectors for NKM
x_vec = zeros(1,Q+burnin);
pi_vec = zeros(1,Q+burnin);
sq_vec = zeros(1,Q+burnin);
i_vec = zeros(1,Q+burnin);
r_vec = zeros(1,Q+burnin);
dq_vec = zeros(1,Q+burnin);

Ex_vec = zeros(1,Q+burnin); % Expectation: E_q[q+1] is located at position q in this vector
Ex_tar_vec = zeros(1,Q+burnin);
Ex_hab_vec = zeros(1,Q+burnin);
Ex_ext_vec = zeros(1,Q+burnin);
Epi_vec = zeros(1,Q+burnin);
Epi_tar_vec = zeros(1,Q+burnin);
Epi_rof_vec = zeros(1,Q+burnin);
Epi_ext_vec = zeros(1,Q+burnin);
Es_vec = zeros(1,Q+burnin);
Es_tar_vec = zeros(1,Q+burnin);
Es_sta_vec = zeros(1,Q+burnin);
Es_ext_vec = zeros(1,Q+burnin);

A_x_tar_vec = zeros(1,Q+burnin); % Attractivity of optimist forcasting of x
A_x_hab_vec = zeros(1,Q+burnin);
A_x_ext_vec = zeros(1,Q+burnin);
A_pi_tar_vec = zeros(1,Q+burnin); % Attractivity of extrapolators forcasting of pi
A_pi_rof_vec = zeros(1,Q+burnin);
A_pi_ext_vec = zeros(1,Q+burnin);
A_s_tar_vec = zeros(1,Q+burnin); % Attractivity of extrapolators forcasting of pi
A_s_sta_vec = zeros(1,Q+burnin);
A_s_ext_vec = zeros(1,Q+burnin);

W_x_tar_vec = zeros(1,Q+burnin);
W_x_hab_vec = zeros(1,Q+burnin);
W_x_ext_vec = zeros(1,Q+burnin);
W_pi_tar_vec = zeros(1,Q+burnin);
W_pi_rof_vec = zeros(1,Q+burnin);
W_pi_ext_vec = zeros(1,Q+burnin);
W_s_tar_vec = zeros(1,Q+burnin);
W_s_sta_vec = zeros(1,Q+burnin);
W_s_ext_vec = zeros(1,Q+burnin);

% Data Vectors for ABC
st_vec = ones(1,T+1)*f_bar; %One longer for the last price adjust
f_vec = ones(1,T)*f_bar;
DF_vec = zeros(1,T);
DC_vec = zeros(1,T);
AF_vec = zeros(1,T);
AC_vec = zeros(1,T);
WF_vec = zeros(1,T);
WC_vec = zeros(1,T);
ProfitTaxless_F_vec = zeros(1,T);
ProfitTaxless_C_vec = zeros(1,T);

TaxRevenue_quarterly = zeros(1,Q+burnin);
TaxRevenue = zeros(1,T);

%% Generate data
noise_nkm = randn(3,Q+burnin)*sig_eps; % contains [epsilon_i; epsilon_x; epsilon_pi]
%noise_sm = randn(3,(Q+burnin)*64) .* ([var_s;var_C;var_F]*ones(1,(Q+burnin)*64)); % contains [epsilon_s; epsilon_C; epsilon_F]

if Turn_Noise_Off_in_q>0 && Turn_Noise_Off_in_q<Q
noise_nkm(:,burnin+Turn_Noise_Off_in_q:end) = 0;
%noise_sm(:,(burnin+Turn_Noise_Off_in_q)*64:end) = 0;
end
for q = 3:Q+burnin % Loop through quarters

if q==5+burnin % Perform the shock: Make people more pessimistic (Caution, A is negative)
if (Shock==4)
tau = 0.01;
end
end

if q==Turn_Noise_Off_in_q
var_s = 0;
var_C = 0;
var_F = 0;
end

for t = ((q-1)*64+1):q*64 % Loop through days

f_vec(t) = h*x_vec(q-1);

ProfitTaxless_F_vec(t) = (exp(st_vec(t))-exp(st_vec(t-1)))*DF_vec(t-2);
ProfitTaxless_C_vec(t) = (exp(st_vec(t))-exp(st_vec(t-1)))*DC_vec(t-2);

% Attractivity
if strcmp(tax_type,'FTT')
AF_vec(t) = ProfitTaxless_F_vec(t) - tau*(exp(st_vec(t))+exp(st_vec(t-1)))*abs(DF_vec(t-2)) + d*AF_vec(t-1);
AC_vec(t) = ProfitTaxless_C_vec(t) - tau*(exp(st_vec(t))+exp(st_vec(t-1)))*abs(DC_vec(t-2)) + d*AC_vec(t-1);
TaxRevenue(t) = tau*(exp(st_vec(t))+exp(st_vec(t-1))) * (abs(DF_vec(t-2))*WF_vec(t-2)+abs(DC_vec(t-2))*WC_vec(t-2));
elseif strcmp(tax_type,'FAT')
AF_vec(t) = ProfitTaxless_F_vec(t) - (ProfitTaxless_F_vec(t) > 0)*tau*ProfitTaxless_F_vec(t) + d*AF_vec(t-1);
AC_vec(t) = ProfitTaxless_C_vec(t) - (ProfitTaxless_C_vec(t) > 0)*tau*ProfitTaxless_C_vec(t) + d*AC_vec(t-1);
TaxRevenue(t) = (ProfitTaxless_F_vec(t) > 0)*tau*ProfitTaxless_F_vec(t)*WF_vec(t-2) + (ProfitTaxless_C_vec(t) > 0)*tau*ProfitTaxless_C_vec(t)*WC_vec(t-2);
elseif strcmp(tax_type,'FAT_progressive')
FAT_rate_F = (ProfitTaxless_F_vec(t) > 0) * min(1, tau * ProfitTaxless_F_vec(t)^2 / 0.0015^2 * (ProfitTaxless_F_vec(t) > 0) );
FAT_rate_C = (ProfitTaxless_C_vec(t) > 0) * min(1, tau * ProfitTaxless_C_vec(t)^2 / 0.0015^2 * (ProfitTaxless_C_vec(t) > 0) );
AF_vec(t) = ProfitTaxless_F_vec(t) - FAT_rate_F*ProfitTaxless_F_vec(t) + d*AF_vec(t-1);
AC_vec(t) = ProfitTaxless_C_vec(t) - FAT_rate_C*ProfitTaxless_C_vec(t) + d*AC_vec(t-1);
TaxRevenue(t) = (ProfitTaxless_F_vec(t) > 0)*FAT_rate_F*ProfitTaxless_F_vec(t)*WF_vec(t-2) + (ProfitTaxless_C_vec(t) > 0)*FAT_rate_C*ProfitTaxless_C_vec(t)*WC_vec(t-2);
end

% Weights
WF_vec(t) = exp(e*AF_vec(t)) / (exp(e*AF_vec(t))+exp(e*AC_vec(t))+1);
WC_vec(t) = exp(e*AC_vec(t)) / (exp(e*AF_vec(t))+exp(e*AC_vec(t))+1);

% Demands
DF_vec(t) = c*(f_vec(t)-st_vec(t)) + randn*var_F;
DC_vec(t) = b*(st_vec(t)-st_vec(t-1)) + randn*var_C;

% Stock price
st_vec(t+1) = st_vec(t) + a*(WF_vec(t)*DF_vec(t)+WC_vec(t)*DC_vec(t)+k/64*(dq_vec(q-1)-dq_vec(q-2)+del_d_s*st_vec(t))) + randn*var_s;

end

TaxRevenue_quarterly(q) = sum(TaxRevenue(((q-1)*64+1):q*64));
sq_vec(q) = st_vec((q-1)*64+1:q*64) * st_sq_weights; % Calculate relevant mean stock price for NKM model
%TaxRevenue_quarterly(q) = 0;
%sq_vec(q) = 0;

% Attractivity
A_x_tar_vec(q) = -(x_vec(q-1) -Ex_tar_vec(q-2))^2 + zeta*A_x_tar_vec(q-1);
A_x_hab_vec(q) = -(x_vec(q-1) -Ex_hab_vec(q-2))^2 + zeta*A_x_hab_vec(q-1);
A_x_ext_vec(q) = -(x_vec(q-1) -Ex_ext_vec(q-2))^2 + zeta*A_x_ext_vec(q-1);
A_pi_tar_vec(q) = -(pi_vec(q-1)-Epi_tar_vec(q-2))^2 + zeta*A_pi_tar_vec(q-1);
A_pi_rof_vec(q) = -(pi_vec(q-1)-Epi_rof_vec(q-2))^2 + zeta*A_pi_rof_vec(q-1);
A_pi_ext_vec(q) = -(pi_vec(q-1)-Epi_ext_vec(q-2))^2 + zeta*A_pi_ext_vec(q-1);
A_s_tar_vec(q) = -(sq_vec(q-1)-Es_tar_vec(q-2))^2 + zeta*A_s_tar_vec(q-1);
A_s_sta_vec(q) = -(sq_vec(q-1)-Es_sta_vec(q-2))^2 + zeta*A_s_sta_vec(q-1);
A_s_ext_vec(q) = -(sq_vec(q-1)-Es_ext_vec(q-2))^2 + zeta*A_s_ext_vec(q-1);

% Weights
W_x_tar_vec(q) = exp(phi*A_x_tar_vec(q)) / (exp(phi*A_x_tar_vec(q)) +exp(phi*A_x_hab_vec(q)) +exp(phi*A_x_ext_vec(q)));
W_x_hab_vec(q) = exp(phi*A_x_hab_vec(q)) / (exp(phi*A_x_tar_vec(q)) +exp(phi*A_x_hab_vec(q)) +exp(phi*A_x_ext_vec(q)));
W_x_ext_vec(q) = exp(phi*A_x_ext_vec(q)) / (exp(phi*A_x_tar_vec(q)) +exp(phi*A_x_hab_vec(q)) +exp(phi*A_x_ext_vec(q)));
if sum(isnan([W_x_tar_vec(q) W_x_hab_vec(q) W_x_ext_vec(q)])) > 0 % if all weights are very small, a division by zero would be created below
W_x_tar_vec(q) = 0;
W_x_hab_vec(q) = 0;
W_x_ext_vec(q) = 0;
if A_x_tar_vec(q) > A_x_hab_vec(q) && A_x_tar_vec(q) > A_x_ext_vec(q)
W_x_tar_vec(q) = 1;
elseif A_x_hab_vec(q) > A_x_tar_vec(q) && A_x_hab_vec(q) > A_x_ext_vec(q)
W_x_hab_vec(q) = 1;
else
W_x_ext_vec(q) = 1;
end
end
W_pi_tar_vec(q) = exp(phi*A_pi_tar_vec(q)) / (exp(phi*A_pi_tar_vec(q))+exp(phi*A_pi_rof_vec(q))+exp(phi*A_pi_ext_vec(q)));
W_pi_rof_vec(q) = exp(phi*A_pi_rof_vec(q)) / (exp(phi*A_pi_tar_vec(q))+exp(phi*A_pi_rof_vec(q))+exp(phi*A_pi_ext_vec(q)));
W_pi_ext_vec(q) = exp(phi*A_pi_ext_vec(q)) / (exp(phi*A_pi_tar_vec(q))+exp(phi*A_pi_rof_vec(q))+exp(phi*A_pi_ext_vec(q)));
if sum(isnan([W_pi_tar_vec(q) W_pi_rof_vec(q) W_pi_ext_vec(q)])) > 0 % if all weights are very small, a division by zero would be created below
W_pi_tar_vec(q) = 0;
W_pi_rof_vec(q) = 0;
W_pi_ext_vec(q) = 0;
if A_pi_tar_vec(q) > A_pi_rof_vec(q) && A_pi_tar_vec(q) > A_pi_ext_vec(q)
W_pi_tar_vec(q) = 1;
elseif A_pi_rof_vec(q) > A_pi_tar_vec(q) && A_pi_rof_vec(q) > A_pi_ext_vec(q)
W_pi_rof_vec(q) = 1;
else
W_pi_ext_vec(q) = 1;
end
end
W_s_tar_vec(q) = exp(phi*A_s_tar_vec(q)) / (exp(phi*A_s_tar_vec(q))+exp(phi*A_s_sta_vec(q))+exp(phi*A_s_ext_vec(q)));
W_s_sta_vec(q) = exp(phi*A_s_sta_vec(q)) / (exp(phi*A_s_tar_vec(q))+exp(phi*A_s_sta_vec(q))+exp(phi*A_s_ext_vec(q)));
W_s_ext_vec(q) = exp(phi*A_s_ext_vec(q)) / (exp(phi*A_s_tar_vec(q))+exp(phi*A_s_sta_vec(q))+exp(phi*A_s_ext_vec(q)));

% time dependend system variables
A_q = [sig-(sig-(1-eta)*del_x)*W_x_ext_vec(q)*(1+alp_x)-sig*c_1*W_s_tar_vec(q)*h -(1-(1-eta)*del_pi-sig*c_1)*W_pi_ext_vec(q)*(1+alp_pi) ;
-gam 1-bet*W_pi_ext_vec(q)*(1+alp_pi)];
C_q = [(sig-(1-eta)*del_x)*(W_x_hab_vec(q)-W_x_ext_vec(q)*alp_x) (1-(1-eta)*del_pi-sig*c_1)*(W_pi_rof_vec(q)-W_pi_ext_vec(q)*alp_pi) ;
0 bet*(W_pi_rof_vec(q)-W_pi_ext_vec(q)*alp_pi)];
D_q = [c_1*sig*(W_s_ext_vec(q)*(1+alp_s)-1) ; -kap];
E_q = [c_1*sig*(W_s_sta_vec(q)-alp_s*W_s_ext_vec(q)) ; 0];

res = inv(A_q)*C_q*[x_vec(q-1); pi_vec(q-1)] + inv(A_q)*D_q*sq_vec(q) + inv(A_q)*E_q*sq_vec(q-1) + inv(A_q)*[-eta;0]*i_vec(q-1) + inv(A_q)*[ sig*noise_nkm(2,q)-noise_nkm(1,q) ; noise_nkm(3,q) ];

if q==5+burnin % Perform the shock
if (Shock==1) % cost shock
res = res + inv(A_q)*[ 0-0/sig ; 1 ];
elseif (Shock==3) % interest rate shock
res = res + inv(A_q)*[ 0-1/sig ; 0 ];
end
end

% Save results in other vector
x_vec(q) = res(1);
pi_vec(q) = res(2);

% Expectations for q+1
Ex_tar_vec(q) = 0;
Ex_hab_vec(q) = x_vec(q-1);
Ex_ext_vec(q) = x_vec(q) + alp_x*(x_vec(q)-x_vec(q-1));
Ex_vec(q) = W_x_tar_vec(q)*Ex_tar_vec(q) + W_x_hab_vec(q)*Ex_hab_vec(q) + W_x_ext_vec(q)*Ex_ext_vec(q); % market expectation of output gap
Epi_tar_vec(q) = 0;
Epi_rof_vec(q) = pi_vec(q-1);
Epi_ext_vec(q) = pi_vec(q) + alp_pi*(pi_vec(q)-pi_vec(q-1));
Epi_vec(q) = W_pi_tar_vec(q)*Epi_tar_vec(q) + W_pi_rof_vec(q)*Epi_rof_vec(q) + W_pi_ext_vec(q)*Epi_ext_vec(q); % market expectation of output gap
Es_tar_vec(q) = 0;
Es_sta_vec(q) = sq_vec(q-1);
Es_ext_vec(q) = sq_vec(q) + alp_s*(sq_vec(q)-sq_vec(q-1));
Es_vec(q) = W_s_tar_vec(q)*Es_tar_vec(q) + W_s_sta_vec(q)*Es_sta_vec(q) + W_s_ext_vec(q)*Es_ext_vec(q); % market expectation of output gap

% interest rate
i_vec(q) = del_pi*Epi_vec(q) + del_x*Ex_vec(q) + noise_nkm(1,q);
r_vec(q) = i_vec(q) - Epi_vec(q);

% stock demand of households
dq_vec(q) = x_vec(q) -c_3*(sq_vec(q)-sum(pi_vec(1:q))) - c_4*i_vec(q);

end

%% Check if s exploded
if isnan(st_vec(length(st_vec))) || st_vec(end) > 10000
if silent == false
disp('Exploded! :-( ');
end
exploded = 1;
end

%% Put all results into "result" variable
result.x = x_vec(burnin+1:(Q+burnin));
result.pi = pi_vec(burnin+1:(Q+burnin));
result.sq = sq_vec(burnin+1:(Q+burnin));
result.i = i_vec(burnin+1:(Q+burnin));
result.r = r_vec(burnin+1:(Q+burnin));
result.dq = dq_vec(burnin+1:(Q+burnin));
result.Ex = Ex_vec(burnin+1:(Q+burnin));
result.Ex_tar = Ex_tar_vec(burnin+1:(Q+burnin));
result.Ex_hab = Ex_hab_vec(burnin+1:(Q+burnin));
result.Ex_ext = Ex_ext_vec(burnin+1:(Q+burnin));
result.W_x_tar = W_x_tar_vec(burnin+1:(Q+burnin));
result.W_x_hab = W_x_hab_vec(burnin+1:(Q+burnin));
result.W_x_ext = W_x_ext_vec(burnin+1:(Q+burnin));
result.Epi = Epi_vec(burnin+1:(Q+burnin));
result.Epi_tar = Epi_tar_vec(burnin+1:(Q+burnin));
result.Epi_rof = Epi_rof_vec(burnin+1:(Q+burnin));
result.Epi_ext = Epi_ext_vec(burnin+1:(Q+burnin));
result.W_pi_tar = W_pi_tar_vec(burnin+1:(Q+burnin));
result.W_pi_rof = W_pi_rof_vec(burnin+1:(Q+burnin));
result.W_pi_ext = W_pi_ext_vec(burnin+1:(Q+burnin));

result.st = st_vec(burnin*64+1:(Q+burnin)*64);
result.f = f_vec(burnin*64+1:(Q+burnin)*64);
result.WF = WF_vec(burnin*64+1:(Q+burnin)*64);
result.WC = WC_vec(burnin*64+1:(Q+burnin)*64);
result.ProfitTaxless_F = ProfitTaxless_F_vec(burnin*64+1:(Q+burnin)*64);
result.ProfitTaxless_C = ProfitTaxless_C_vec(burnin*64+1:(Q+burnin)*64);

result.TaxRevenue_quarterly = TaxRevenue_quarterly(burnin+1:(Q+burnin));
result.TaxRevenue = TaxRevenue(burnin*64+1:(Q+burnin)*64);

result.exploded = exploded;

Funktion ohne Link?

Das Problem liegt hier: Ich kann diesen Code nicht nachvollziehen. Es handelt sich um ein Optimierungsporblem:

Code:

function [ LossVal, vol_pi, vol_x,vol_s ] = Loss_Fun( data, w_pi, w_x )

vol_pi = (mean(data.pi.^2));
vol_x = (mean(data.x.^2));
vol_s = (mean(data.st.^2));

LossVal = w_pi * vol_pi ...
+ w_x * vol_x ...
+ w_s * vol_s;

w_pi = 0.5 ; %1;
w_x = 0.25; %0.5;

end

Funktion ohne Link?

Code:

function [ LossVal, vol_pi, vol_x, vol_s ] = Eval_Loss_Fun( optim_par, fix_par )

% Set the data generating process
global DGP

% Check if results already exist
file_name_BerechnungsSpeicher = [num2str(optim_par,20) ' ' num2str(fix_par,20)];
for i = 1:5
file_name_BerechnungsSpeicher = strrep(file_name_BerechnungsSpeicher,' ',' ');
file_name_BerechnungsSpeicher = strrep(file_name_BerechnungsSpeicher,' ',' ');
file_name_BerechnungsSpeicher = strrep(file_name_BerechnungsSpeicher,' ',' ');
end
warning off; mkdir(['Berechnungsspeicher/' DGP]); warning on;
file_name_BerechnungsSpeicher = ['Berechnungsspeicher/' DGP '/' strrep(file_name_BerechnungsSpeicher,' ','_') '.mat'];
%delete(file_name_BerechnungsSpeicher);

if exist(file_name_BerechnungsSpeicher, 'file') == 0

del_pi = optim_par(1);
del_x = optim_par(2);

if length(optim_par) >= 3
del_d_s = optim_par(3);
else
del_d_s = NaN;
end

% interaction parameter
k = fix_par(1); %0.2;
kap = fix_par(2); %0.1;
c_1 = fix_par(3); %0.2;
h = fix_par(4); %0.5;

% Weights for Loss Function
w_pi = fix_par(5); %1;
w_x = fix_par(6); %0.5;
w_s = fix_par(7); %0;

if length(fix_par) >= 8
eta = fix_par(8); %0;
else
eta = 0;
end

no_of_seeds = 100;

seeds = 100:100:100*no_of_seeds;
LossVal_vec = zeros(size(seeds));
vol_pi_vec = zeros(size(seeds));
vol_x_vec = zeros(size(seeds));
vol_s_vec = zeros(size(seeds));
for seed_pos = 1:no_of_seeds
if isnan(del_d_s)
eval(['data = ' DGP '(500, ' num2str(seeds(seed_pos)) ', ''del_pi'',' num2str(del_pi) ', ''del_x'',' num2str(del_x) ', ''k'',' num2str(k) ', ''kap'',' num2str(kap) ', ''c_1'',' num2str(c_1) ', ''h'',' num2str(h) ', ''eta'',' num2str(eta) ', ''silent'');']);
%data = GenerateDynamics_Erw2(500, seeds(seed_pos), 'del_pi',del_pi, 'del_x',del_x, 'k',k, 'kap',kap, 'c_1',c_1, 'h',h, 'silent');
else
eval(['data = ' DGP '(500, ' num2str(seeds(seed_pos)) ', ''del_pi'',' num2str(del_pi) ', ''del_x'',' num2str(del_x) ', ''del_d_s'', ' num2str(del_d_s) ', ''k'',' num2str(k) ', ''kap'',' num2str(kap) ', ''c_1'',' num2str(c_1) ', ''h'',' num2str(h) ', ''eta'',' num2str(eta) ', ''silent'');']);
end
[ LossVal_vec(seed_pos), vol_pi_vec(seed_pos), vol_x_vec(seed_pos), vol_s_vec(seed_pos) ] = Loss_Fun( data, w_pi, w_x, w_s );
%pause(0.01)
end

LossVal = mean(LossVal_vec);
vol_pi = mean(vol_pi_vec);
vol_x = mean(vol_x_vec);
vol_s = mean(vol_s_vec);

if isnan(LossVal)
LossVal = inf;
end

% Save Loss_Value to hard disk
save(file_name_BerechnungsSpeicher, 'LossVal', 'vol_pi', 'vol_x', 'vol_s');

else

% Lode previously saved result from hard disk
load(file_name_BerechnungsSpeicher);
pause(0.01);

end

end

Funktion ohne Link?

Ich stehe komplett auf dem Schlauch hier. Zuerst wird die Lossfunktion definiert, das erkenne ich nicht. Aber was dann danach passiert, verstehe ich nicht. Wie man an dem unteren Teil des Codes erkennen kann, sollen optimale parameter gefunden werden, um die Lossfunktion so klein wie möglich zu halten.

Lisa89

Themenstarter

Forum-Anfänger


	Beiträge: 33

	Anmeldedatum: 13.12.18

	Wohnort: ---

	Version: ---

Verfasst am: 27.11.2019, 15:42 Titel:

Ich habe einen Ansatz gefunden, um das Optimierungsproblem zu lösen und zwar mit dem Nelder Mead Algorithmus. Aber dafür bräuchte ich Hilfe. Ich hab diese Beschreibung im Internet gefunden und das ist genau das, was ich machen muss:

The loss_search_ ﬁles start the Nelder-Mead algorithm that optimizes the coefﬁcients of the Taylor rule according to the welfare loss obained by the functions loss_.

Taylor Rule ist in meinem Fall:

i= delta_pi*pi +delta_x*x + ut

und meine Lossfunktion ist:
lim L= 1/2 (Var pi + lambda Var xt)

die koeffizienten sind delta_pi and delta_x


Einstellungen und Berechtigungen
Beiträge der letzten Zeit anzeigen:	Du kannst Beiträge in dieses Forum schreiben. Du kannst auf Beiträge in diesem Forum antworten. Du kannst deine Beiträge in diesem Forum nicht bearbeiten. Du kannst deine Beiträge in diesem Forum nicht löschen. Du kannst an Umfragen in diesem Forum nicht mitmachen. Du kannst Dateien in diesem Forum posten Du kannst Dateien in diesem Forum herunterladen

Impressum | Nutzungsbedingungen | Datenschutz | FAQ |

RSS

Hosted by:

Copyright © 2007 - 2024 goMatlab.de | Dies ist keine offizielle Website der Firma The Mathworks

MATLAB, Simulink, Stateflow, Handle Graphics, Real-Time Workshop, SimBiology, SimHydraulics, SimEvents, and xPC TargetBox are registered trademarks and The MathWorks, the L-shaped membrane logo, and Embedded MATLAB are trademarks of The MathWorks, Inc.