www.gusucode.com > EasyKrig_V3.0工具箱matlab源码程序 > EasyKrig_V3.0/visualization/cross_validation.m
function out=cross_validation(opt1,opt2)
% function cross_validation(opt) performs cross validation computation
% opt1 = 1 Q1 cross validation
% 2 Q2 cross validation
% 3 double kriging
% 4 Jackknife
% opt2 = 1 Compute if it has not computed before
% 2 Re-compute use the current variogram parameter settings
%%
%% Kriging Software Package version 3.0, May 1, 2004
%% Copyright (c) 1999, 2001, 2004, property of Dezhang Chu and Woods Hole Oceanographic
%% Institution. All Rights Reserved.
global data para hdl
out=1;
if ~isfield(para,'vario')
message(1,'Need to performing variogram first!!');
out=-1;
return
elseif ~isfield(para.vario,'model')
message(1,'Need to performing kriging first!!');
out=-1;
return
end
vmod=para.vario.model;
if para.vario.corr == 1
vmod = -vmod;
end
model_para=[para.vario.nugt para.vario.sill para.vario.lscl para.vario.powr para.vario.hole];
%% kriging parameters
mod_opt=para.krig.model;
srad=para.krig.srad;
kmin=para.krig.kmin;
kmax=para.krig.kmax;
blk_opt=para.krig.scheme;
blknx=para.krig.blk_nx;
blkny=para.krig.blk_ny;
elim=para.krig.elim;
dx=para.krig.dx;
dy=para.krig.dy;
EPS=para.krig.eps;
var=data.in.tv;
x=data.in.x;
y=data.in.y;
if data.in.dim == 3
dz=para.krig.dz;
z=data.in.z;
indx_nan=find(isnan(x)|isnan(y)|isnan(z));
if ~isempty(indx_nan)
z(indx_nan)=[];
end
else
indx_nan=find(isnan(x)|isnan(y));
end
if ~isempty(indx_nan)
indx_orig=[1:length(x)]';
data.out.krig.sta=setdiff(indx_orig,indx_nan);
x(indx_nan)=[];
y(indx_nan)=[];
var(indx_nan)=[];
else
data.out.krig.sta=1:length(x);
end
msg1='Cross-Validation in progress, ...';
msg2=' ';
nmin=8; % standard for displaying time elapse
if ~isfield(para.krig,'model')
krig_mod=2; % default kriging method: ordinary Kriging
else
krig_mod=para.krig.model;
end
n=length(x);
tic
if opt1 <= 2
% % Q-1
if para.dispkrig.Qcheck == 0 | opt2 == 1
sigma0=sqrt(para.vario.c0);
n=length(x);
if n >= nmin hmsg=message(2,msg1,' ');drawnow;end
nv=n-mod_opt;
nv_cnt=max(1,round(0.01*nv));
for i=mod_opt+1:n
j=i-mod_opt;
if rem(j,nv_cnt) == 0
percent=sprintf(' %g percent done in %g sec.', floor(100*j/nv),round(toc));
if n >= nmin message(4,percent,' ',hmsg);drawnow; end
end
if data.in.dim == 2
xc=x(i);yc=y(i);
xi=x(1:i-1);yi=y(1:i-1);var_in=var(1:i-1);
if krig_mod == 1 % remove the mean for simple kriging
var_mean=mean(var_in);
var_in=var_in-var_mean;
end
% para.krig.kmin=mod_opt;
% para.krig.kmax=length(xi);
% para.krig.srad=0.3;
[var_out,err_krig]=krig2d(xc,yc,xi,yi,var_in,model_para);
if para.krig.model == 1 % add in the mean for simple kriging
var_out=var_out+var_mean;
end
else % 3-D kriging
xc=x(i);yc=y(i);zc=z(i);
xi=x(1:i-1);yi=y(1:i-1);zi=z(1:i-1);var_in=var(1:i-1);
if krig_mod == 1 % remove the mean for simple kriging
var_mean=mean(var_in);
var_in=var_in-var_mean;
end
[var_out,err_krig]=krig3d(xc,yc,zc,xi,yi,zi,var_in,model_para);
if para.krig.model == 1 % add in the mean for simple kriging
var_out=var_out+var_mean;
end
end
if err_krig < 0 err_krig=nan;end
ek(i-mod_opt)=(var(i)-var_out)/(sigma0*sqrt(err_krig));
errk(i-mod_opt)=err_krig;
end
data.out.krig.q1=mean_nan(ek);
data.out.krig.q2=mean_nan(ek.^2);
data.out.krig.ek=ek;
if n > nmin close(hmsg);end
else
data.out.krig.q1=mean_nan(data.out.krig.ek);
data.out.krig.q2=mean_nan(data.out.krig.ek.^2);
end
para.dispkrig.Qcheck=1;
set(hdl.validation.recompute,'enable','on');
elseif opt1 == 3 % double kriging - default
if para.dispkrig.Dblkrig == 0 | opt2 == 1
n=length(x);
if n >= nmin hmsg=message(2,msg1,' ');end
if data.in.dim == 2
xc=x;yc=y;
xi=data.out.krig.xp;
yi=data.out.krig.yp;
var_in=data.out.krig.var;
if krig_mod == 1 % remove the mean for simple kriging
var_mean=mean(var_in);
var_in=var_in-var_mean;
end
[var_out,err_krig]=krig2d(xc,yc,xi,yi,var_in,model_para,hmsg);
if para.krig.model == 1 % add in the mean for simple kriging
var_out=var_out+var_mean;
end
else
xc=x;yc=y;zc=z;
xi=data.out.krig.xp;
yi=data.out.krig.yp;
zi=data.out.krig.zp;
var_in=data.out.krig.var;
if krig_mod == 1 % remove the mean for simple kriging
var_mean=mean(var_in);
var_in=var_in-var_mean;
end
[var_out,err_krig]=krig3d(xc,yc,zc,xi,yi,zi,var_in,model_para,hmsg);
if para.krig.model == 1 % add in the mean for simple kriging
var_out=var_out+var_mean;
end
end
data.out.krig.Is=var_out(:);
if n > nmin close(hmsg);end
end
para.dispkrig.Dblkrig=1;
set(hdl.validation.recompute,'enable','on');
elseif opt1 == 4 % jackknife
if para.dispkrig.JKcheck == 0 | opt2 == 1
var=data.in.tv;
n=length(x);
if n >= nmin hmsg=message(2,msg1,' ');end
Ijk(1)=var(1);
nv=n-mod_opt;
nv_cnt=max(1,round(0.01*nv));
for i=2:n
j=i-mod_opt;
if rem(j,nv_cnt) == 0
percent=sprintf(' %g percent done in %g sec.', floor(100*j/nv),round(toc));
if n >= nmin message(4,percent,' ',hmsg);drawnow; end
end
if data.in.dim == 2
xc=x(i);yc=y(i);
xi=[x(1:i-1); x(i+1:n)];yi=[y(1:i-1); y(i+1:n)];var_in=[var(1:i-1); var(i+1:n)];
if krig_mod == 1 % remove the mean for simple kriging
var_mean=mean(var_in);
var_in=var_in-var_mean;
end
[var_out,err_krig]=krig2d(xc,yc,xi,yi,var_in,model_para);
if para.krig.model == 1 % add in the mean for simple kriging
var_out=var_out+var_mean;
end
else
xc=x(i);yc=y(i);zc=z(i);
xi=[x(1:i-1); x(i+1:n)];yi=[y(1:i-1); y(i+1:n)];zi=[z(1:i-1); z(i+1:n)];var_in=[var(1:i-1); var(i+1:n)];
if krig_mod == 1 % remove the mean for simple kriging
var_mean=mean(var_in);
var_in=var_in-var_mean;
end
[var_out,err_krig]=krig3d(xc,yc,zc,xi,yi,zi,var_in,model_para);
if para.krig.model == 1 % add in the mean for simple kriging
var_out=var_out+var_mean;
end
end
Ijk(i)=var_out;
end
data.out.krig.Ijk=Ijk(:);
if n > nmin close(hmsg);end
end
para.dispkrig.JKcheck=1;
set(hdl.validation.recompute,'enable','on');
end