stepan
/
dynare
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Improve tractability of McMCDiagnostics_core.m by adding header and comments. Also uses more intuitive variable names

time-shift
Johannes Pfeifer 2014-10-09 10:57:28 +02:00
parent 6bbfa2fa6d
commit 1d68ed8da6
1 changed files with 44 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

72
matlab/McMCDiagnostics_core.m
View File

@ -1,23 +1,39 @@
function myoutput = McMCDiagnostics_core(myinputs,fpar,npar,whoiam, ThisMatlab)
% function myoutput = McMCDiagnostics_core(myinputs,fpar,npar,whoiam, ThisMatlab)
% Computes the Brooks/Gelman (1998) convergence diagnostics, both the
% parameteric and the non-parameteric versions
%
% PARALLEL CONTEXT
% Core functionality for MCMC Diagnostics, which can be parallelized.
% See also the comment in random_walk_metropolis_hastings_core.m funtion.
%
%
% INPUTS
% See See the comment in random_walk_metropolis_hastings_core.m funtion.
% OUTPUTS
% o myoutput [struc]
% Contained UDIAG.
% Contains:
% - UDIAG [by 6] double 1st column: length of total sequence interval
% 2nd column: sum of length of within sequence intervals; used to compute mean length of within sequence intervals
% 3nd column: within sequence variance
% 4nd column: sum of within sequence variances; used to compute mean within sequence variances
% 5nd column: within sequence kurtosis
% 6nd column: sum of within sequence kurtoses; used to compute mean within sequence kurtoses
% Averaging to compute mean moments is done in
% McMCDiagnostics
%
% ALGORITHM
% Portion of McMCDiagnostics.m function.
% Computes part of the convergence diagnostics, the rest is computed in McMCDiagnostics.m .
% The methodology and terminology is based on: Brooks/Gelman (1998): General
% Methods for Monitoring Convergence of Iterative Simulations, Journal of Computational
% and Graphical Statistics, Volume 7, Number 4, Pages 434-455
%
%
% SPECIAL REQUIREMENTS.
% None.
% Copyright (C) 2006-2011 Dynare Team
% Copyright (C) 2006-2014 Dynare Team
%
% This file is part of Dynare.
%
@ -59,7 +75,7 @@ if ~exist('MetropolisFolder'),
MetropolisFolder = CheckPath('metropolis',M_.dname);
end
ALPHA = 0.2; % increase too much with the number of simulations.
ALPHA = 0.2; % percentile for non-parametric statistic
tmp = zeros(NumberOfDraws*nblck,3);
UDIAG = zeros(NumberOfLines,6,npar-fpar+1);
@ -80,7 +96,7 @@ for j=fpar:npar,
else
fprintf(' Parameter %d... ',j);
end
for b = 1:nblck
for b = 1:nblck %load draws from different chains into 1 matrix
startline = 0;
for n = 1:NumberOfMcFilesPerBlock
load([MetropolisFolder '/' M_.fname '_mh',int2str(n),'_blck' int2str(b) '.mat'],'x2');
@ -89,29 +105,29 @@ for j=fpar:npar,
startline = startline + nx2;
end
end
tmp(:,2) = kron(transpose(1:nblck),ones(NumberOfDraws,1));
tmp(:,3) = kron(ones(nblck,1),time');
tmp = sortrows(tmp,1);
ligne = 0;
for iter = Origin:StepSize:NumberOfDraws
ligne = ligne+1;
linea = ceil(mh_drop*iter);
n = iter-linea+1;
cinf = round(n*ALPHA/2);
csup = round(n*(1-ALPHA/2));
CINF = round(nblck*n*ALPHA/2);
CSUP = round(nblck*n*(1-ALPHA/2));
temp = tmp(find((tmp(:,3)>=linea) & (tmp(:,3)<=iter)),1:2);
UDIAG(ligne,1,j-fpar+1) = temp(CSUP,1)-temp(CINF,1);
moyenne = mean(temp(:,1));%% Pooled mean.
UDIAG(ligne,3,j-fpar+1) = sum((temp(:,1)-moyenne).^2)/(nblck*n-1);
UDIAG(ligne,5,j-fpar+1) = sum(abs(temp(:,1)-moyenne).^3)/(nblck*n-1);
tmp(:,2) = kron(transpose(1:nblck),ones(NumberOfDraws,1)); %add info about chain associated with draw into 2nd column
tmp(:,3) = kron(ones(nblck,1),time'); %add timeline for draws to third column
tmp = sortrows(tmp,1); %sort draws according to size for non-parametric percentile computation
window_iter = 0;
for iter = Origin:StepSize:NumberOfDraws %begin of window
window_iter = window_iter+1;
linea = ceil(mh_drop*iter); %compute first non-discarded draw; drops fraction of sample at each iteration for computational efficiency, see Brooks/Gelman (1998), p.438
n = iter-linea+1; %number of draws from each block in current batch
cinf = round(n*ALPHA/2); %lower bound for alpha percentile of within series
csup = round(n*(1-ALPHA/2)); %upper bound for alpha percentile of within series
CINF = round(nblck*n*ALPHA/2); %lower bound for alpha percentile of pooled series
CSUP = round(nblck*n*(1-ALPHA/2)); %upper bound for alpha percentile of pooled series
temp = tmp(find((tmp(:,3)>=linea) & (tmp(:,3)<=iter)),1:2); %extract pooled draws in current batch
UDIAG(window_iter,1,j-fpar+1) = temp(CSUP,1)-temp(CINF,1); %length of total sequence interval
pooled_mean = mean(temp(:,1)); % Pooled mean.
UDIAG(window_iter,3,j-fpar+1) = sum((temp(:,1)-pooled_mean).^2)/(nblck*n-1); %within sequence variance
UDIAG(window_iter,5,j-fpar+1) = sum(abs(temp(:,1)-pooled_mean).^3)/(nblck*n-1); %within sequence third moment
for i=1:nblck
pmet = temp(find(temp(:,2)==i));
UDIAG(ligne,2,j-fpar+1) = UDIAG(ligne,2,j-fpar+1) + pmet(csup,1)-pmet(cinf,1);
moyenne = mean(pmet,1); %% Within mean.
UDIAG(ligne,4,j-fpar+1) = UDIAG(ligne,4,j-fpar+1) + sum((pmet(:,1)-moyenne).^2)/(n-1);
UDIAG(ligne,6,j-fpar+1) = UDIAG(ligne,6,j-fpar+1) + sum(abs(pmet(:,1)-moyenne).^3)/(n-1);
UDIAG(window_iter,2,j-fpar+1) = UDIAG(window_iter,2,j-fpar+1) + pmet(csup,1)-pmet(cinf,1); %sum of length of within sequence intervals; used to compute mean length of within sequence intervals
within_mean = mean(pmet,1); %% Within mean in current chain.
UDIAG(window_iter,4,j-fpar+1) = UDIAG(window_iter,4,j-fpar+1) + sum((pmet(:,1)-within_mean).^2)/(n-1); %sum of within sequence variances; used to compute mean within sequence variances
UDIAG(window_iter,6,j-fpar+1) = UDIAG(window_iter,6,j-fpar+1) + sum(abs(pmet(:,1)-within_mean).^3)/(n-1); %sum of within sequence kurtoses; used to compute mean within sequence kurtoses
end
end
if isoctave