plotgpy1 handles NaN values

Dynamic-Systems-and-GP · Mar 18, 2016 · d9c5c14 · d9c5c14
1 parent b6920d7
commit d9c5c14
Show file tree

Hide file tree

Showing 177 changed files with 23,881 additions and 23,684 deletions.
diff --git a/.gitmodules b/.gitmodules
@@ -1,3 +1,3 @@
-[submodule "gpml-sparse-deriv"]
-	path = gpml-sparse-deriv
-	url = https://github.com/IJS-Systems-and-Control/gpml-sparse-deriv.git
+[submodule "gpml-sparse-deriv"]
+	path = gpml-sparse-deriv
+	url = https://github.com/IJS-Systems-and-Control/gpml-sparse-deriv.git
diff --git a/LICENSE b/LICENSE
diff --git a/README.rst b/README.rst
diff --git a/gpdyn-doc/Makefile b/gpdyn-doc/Makefile
@@ -1,8 +1,8 @@
-all:	readme pdf
-
-readme:
-	pandoc -s --mathjax -S -i manual.tex  -o ../README.rst
-
-pdf:
-	pdflatex manual.tex
-	rm manual.log manual.aux
+all:	readme pdf
+
+readme:
+	pandoc -s --mathjax -S -i manual.tex  -o ../README.rst
+
+pdf:
+	pdflatex manual.tex
+	rm manual.log manual.aux
diff --git a/gpdyn-gp-evaluation/gpx.m b/gpdyn-gp-evaluation/gpx.m
diff --git a/gpdyn-gp-evaluation/simulGPmc.m b/gpdyn-gp-evaluation/simulGPmc.m
@@ -1,125 +1,125 @@
-function [mu, s2, MU, SIG2] = simulGPmc(hyp, inf, meanfunc, cov, lik, input, target, test, lag, Nsamples)
-% simulGPmc - Simulation of the dynamic GP model, where the output variance is
-% propagated using the Monte Carlo method
-%
-%% Syntax
-%  [mu, s2, MU, SIG2] = simulGPmc(hyp, inf, mean, cov, lik, input, target, test,
-%  lag, Nsamples)
-% 
-%% Description
-% Idea: at every time step the output of GP model is approximated with
-% Nsamples samples, which are used as the future inputs of the GP model.
-% Samples are re-used if necessary (ie. y(k-1) for y(k-2) if lag=2 etc.) 
-% Uses routines gpx and gmx_sample. 
-% 
-% Input:
-% * hyp      ... the structure of optimized hyperparameters 
-% * inf      ... the function specifying the inference method 
-% * meanfunc ... the prior mean function
-% * cov      ... the specified covariance function, see help covFun for more info 
-% * lik      ... the likelihood function
-% * input    ... the input part of the training data,  NxD matrix
-% * target   ... the output part of the training data (ie. target), Nx1 vector 
-% * test     ... the input matrix for simulation, kxD vector, see
-%                construct.m for more info  
-% * lag      ... the order of the model (number of used lagged outputs) 
-% * Nsamples ... the number of samples used in algorithm (ie. runs of simulation) 
-% 
-% Output:
-% * mu    ... the mean predicted output 
-% * s2    ... the associated variances (with noise variances)
-% * MU    ... the matrix of all predicted means, kxNsamples
-% * SIG2  ... the associated predicted variances 
-% 
-% See also: 
-% gpx, gmx_sample, simulGPnaive
-% 
-% Examples: 
-% demo_example_gp_simulation
-% 
-%% 
-% * Written by J. Prikryl, November 2010
-% * Based on the work of C.E. Rasmussen, A. Girard, K. Azman. 
-%
-
-% meanfunc ... 'mean' is used as a matlab core function in this file
-
-
-Ndx = 800;
-DSig = 3;
-num_iters = length(test);
-sum_time  = 0;
-
-[N, D] = size(input);
-PDF = zeros(Nsamples,lag);
-
-% Preallocate mu and s2
-mu = zeros ( num_iters, 1 );
-s2 = zeros ( num_iters, 1 );
-
-% 1st step - input is a point
-test_ = test(1,:);
-[mu(1), s2(1), post] = gpx(hyp, inf, meanfunc, cov, lik, input, target, test_);
-
-MU(1,:) = mu(1)*ones(1,Nsamples);
-SIG2(1,:) = s2(1)*ones(1,Nsamples);
-
-pdf = gmx_sample(mu(1),s2(1),Nsamples);
-PDF(:,lag) = pdf;
-% instead of a cycle, create one monstrous test_ matrix
-test_ = repmat( test(2,:), Nsamples, 1 );
-test_(:,lag) = pdf;
-[MU(2,:), SIG2(2,:), post] = gpx(hyp, inf, meanfunc, cov, lik, input, target, test_, post);
-mu(2,1) = mean(MU(2,:));
-s2(2,1) = mean(SIG2(2,:)) + mean((MU(2,:)-mu(2)).^2);
-
-% steps from 3 on ...
-for k=3:num_iters
-
-    if(mod(k,50)==0 || k==3)
-        disp(['simulGPmc, step: ',int2str(k),'/',int2str(length(test))]);    
-    end
-
-
-    if(k>lag)
-        % samples of previous distributions 
-        for jj=1:lag-1
-            PDF(:,jj) = PDF(:,jj+1);
-        end
-
-    else  % k <= lag
-        % part after 'else' not yet tested for lag>=3 
-        col0 = lag-(k-1);
-        PDF(:,1:col0) = repmat(test(k,1:col0),Nsamples,1);
-        for jj=col0+1:lag-1
-            PDF(:,jj) = PDF(:,jj+1);
-        end
-    end
-
-    pdf = gmx_sample(MU(k-1,:),SIG2(k-1,:),Nsamples);
-    PDF(:,lag) = pdf;
-
-    % simulate for all cases, again using the matrix version
-    t0 = tic;
-    test_ = repmat ( test(k,:), Nsamples, 1 );
-    test_(:,1:lag) = PDF;
-    [MU(k,:), SIG2(k,:), post] = gpx(hyp, inf, meanfunc, cov, lik, input, target, test_, post);
-    calltime = toc(t0);
-%     fprintf ( 'gpr_simul() ....... %f sec\n\n', calltime ); % uncomment
-%     if you wish 
-    sum_time = sum_time + calltime;
-
-    % approximate output distribution with gauss - calculate m and v
-    mu(k,1) = mean(MU(k,:));
-    s2(k,1) = mean(SIG2(k,:)) + mean((MU(k,:)-mu(k)).^2);
-
-end
-
-%     fprintf ( 'average time needed for gpr_simul() .... %f sec\n', sum_time/(num_iters-2) );
-%     uncomment if you wish
-return;
-
-
-
-
-
+function [mu, s2, MU, SIG2] = simulGPmc(hyp, inf, meanfunc, cov, lik, input, target, test, lag, Nsamples)
+% simulGPmc - Simulation of the dynamic GP model, where the output variance is
+% propagated using the Monte Carlo method
+%
+%% Syntax
+%  [mu, s2, MU, SIG2] = simulGPmc(hyp, inf, mean, cov, lik, input, target, test,
+%  lag, Nsamples)
+% 
+%% Description
+% Idea: at every time step the output of GP model is approximated with
+% Nsamples samples, which are used as the future inputs of the GP model.
+% Samples are re-used if necessary (ie. y(k-1) for y(k-2) if lag=2 etc.) 
+% Uses routines gpx and gmx_sample. 
+% 
+% Input:
+% * hyp      ... the structure of optimized hyperparameters 
+% * inf      ... the function specifying the inference method 
+% * meanfunc ... the prior mean function
+% * cov      ... the specified covariance function, see help covFun for more info 
+% * lik      ... the likelihood function
+% * input    ... the input part of the training data,  NxD matrix
+% * target   ... the output part of the training data (ie. target), Nx1 vector 
+% * test     ... the input matrix for simulation, kxD vector, see
+%                construct.m for more info  
+% * lag      ... the order of the model (number of used lagged outputs) 
+% * Nsamples ... the number of samples used in algorithm (ie. runs of simulation) 
+% 
+% Output:
+% * mu    ... the mean predicted output 
+% * s2    ... the associated variances (with noise variances)
+% * MU    ... the matrix of all predicted means, kxNsamples
+% * SIG2  ... the associated predicted variances 
+% 
+% See also: 
+% gpx, gmx_sample, simulGPnaive
+% 
+% Examples: 
+% demo_example_gp_simulation
+% 
+%% 
+% * Written by J. Prikryl, November 2010
+% * Based on the work of C.E. Rasmussen, A. Girard, K. Azman. 
+%
+
+% meanfunc ... 'mean' is used as a matlab core function in this file
+
+
+Ndx = 800;
+DSig = 3;
+num_iters = length(test);
+sum_time  = 0;
+
+[N, D] = size(input);
+PDF = zeros(Nsamples,lag);
+
+% Preallocate mu and s2
+mu = zeros ( num_iters, 1 );
+s2 = zeros ( num_iters, 1 );
+
+% 1st step - input is a point
+test_ = test(1,:);
+[mu(1), s2(1), post] = gpx(hyp, inf, meanfunc, cov, lik, input, target, test_);
+
+MU(1,:) = mu(1)*ones(1,Nsamples);
+SIG2(1,:) = s2(1)*ones(1,Nsamples);
+
+pdf = gmx_sample(mu(1),s2(1),Nsamples);
+PDF(:,lag) = pdf;
+% instead of a cycle, create one monstrous test_ matrix
+test_ = repmat( test(2,:), Nsamples, 1 );
+test_(:,lag) = pdf;
+[MU(2,:), SIG2(2,:), post] = gpx(hyp, inf, meanfunc, cov, lik, input, target, test_, post);
+mu(2,1) = mean(MU(2,:));
+s2(2,1) = mean(SIG2(2,:)) + mean((MU(2,:)-mu(2)).^2);
+
+% steps from 3 on ...
+for k=3:num_iters
+
+    if(mod(k,50)==0 || k==3)
+        disp(['simulGPmc, step: ',int2str(k),'/',int2str(length(test))]);    
+    end
+
+
+    if(k>lag)
+        % samples of previous distributions 
+        for jj=1:lag-1
+            PDF(:,jj) = PDF(:,jj+1);
+        end
+
+    else  % k <= lag
+        % part after 'else' not yet tested for lag>=3 
+        col0 = lag-(k-1);
+        PDF(:,1:col0) = repmat(test(k,1:col0),Nsamples,1);
+        for jj=col0+1:lag-1
+            PDF(:,jj) = PDF(:,jj+1);
+        end
+    end
+
+    pdf = gmx_sample(MU(k-1,:),SIG2(k-1,:),Nsamples);
+    PDF(:,lag) = pdf;
+
+    % simulate for all cases, again using the matrix version
+    t0 = tic;
+    test_ = repmat ( test(k,:), Nsamples, 1 );
+    test_(:,1:lag) = PDF;
+    [MU(k,:), SIG2(k,:), post] = gpx(hyp, inf, meanfunc, cov, lik, input, target, test_, post);
+    calltime = toc(t0);
+%     fprintf ( 'gpr_simul() ....... %f sec\n\n', calltime ); % uncomment
+%     if you wish 
+    sum_time = sum_time + calltime;
+
+    % approximate output distribution with gauss - calculate m and v
+    mu(k,1) = mean(MU(k,:));
+    s2(k,1) = mean(SIG2(k,:)) + mean((MU(k,:)-mu(k)).^2);
+
+end
+
+%     fprintf ( 'average time needed for gpr_simul() .... %f sec\n', sum_time/(num_iters-2) );
+%     uncomment if you wish
+return;
+
+
+
+
+