Mercurial > hg > ltpda
view m-toolbox/classes/@ssm/doSimulate.m @ 52:daf4eab1a51e database-connection-manager tip
Fix. Default password should be [] not an empty string
author | Daniele Nicolodi <nicolodi@science.unitn.it> |
---|---|
date | Wed, 07 Dec 2011 17:29:47 +0100 |
parents | f0afece42f48 |
children |
line wrap: on
line source
% DOSIMULATE simulates a discrete ssm with given inputs %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % DESCRIPTION: DOSIMULATE simulates a discrete ssm with given inputs. % % CALL: [x, y, lastX] = doSimulate(SSini, Nsamples, A, Baos, Coutputs, Cstates, Daos, Bnoise, Dnoise, Bcst, Dcst, aos_vect, doTerminate, terminationCond, displayTime, timestep) % % INPUTS: % % OUTPUTS: % % % VERSION: $Id: doSimulate.m,v 1.8 2010/09/08 11:39:20 adrien Exp $ % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % TO DO: Check input aos for the timestep, tsdata, and ssm.timestep % options to be defined (NL case) % add check if one input mach no ssm input variable % allow use of other LTPDA functions to generate white noise function [x, y, lastX] = doSimulate(SSini, Nsamples, A, Baos, Coutputs, Cstates, Daos, Bnoise, Dnoise, Bcst, Dcst, aos_vect, doTerminate, terminationCond, displayTime, timestep, forceComplete) % We do a simple simulate if all these are satisfied: % 1) Bnoise is empty or all zeros % 2) Cstates is empty % 3) Dnoise is empty or all zeros % 4) Bcst is empty or all zeros % 5) Dcst is empty or all zeros % 6) doTerminate is false if (isempty(Bnoise) || all(all(Bnoise==0))) && ... isempty(Cstates) && ... (isempty(Dnoise) || all(all(Dnoise==0))) && ... (isempty(Bcst) || all(all(Bcst==0))) && ... (isempty(Dcst) || all(all(Dcst==0))) && ... ~doTerminate && ... ~forceComplete % do a fast simulation Nstates = numel(SSini); if Nstates >= 100 % except if Matlab is faster [x,y,lastX] = doSimulateSimple(SSini, Nsamples, A, Baos, Coutputs, Daos, aos_vect); else try % call to the mex file x = []; [y,lastX] = ltpda_ssmsim(SSini, A.', Coutputs.', Cstates.', Baos.', Daos.', aos_vect); catch % backup if the mex-file is broken warning('Failed to run mex file ltpda_ssmsim'); [x,y,lastX] = doSimulateSimple(SSini, Nsamples, A, Baos, Coutputs, Daos, aos_vect); end end else % the standard old script, more complete (DC and noise inputs) [x,y,lastX] = doSimulateComplete(SSini, Nsamples, A, Baos, Coutputs, Cstates, Daos, Bnoise, Dnoise, Bcst, Dcst, aos_vect, doTerminate, terminationCond, displayTime, timestep); end end function [x,y,lastX] = doSimulateSimple(lastX, Nsamples, A, Baos, Coutputs, Daos, aos_vect) disp('Running simulate simple...'); % initializing fields x = []; y = zeros(size(Coutputs,1), Nsamples); lastX = A*lastX + Baos*aos_vect(:,1); % time for displaying remaining time % simulation loop for k = 1:Nsamples % computing and storing outputs y(:,k) = Coutputs*lastX + Daos*aos_vect(:,k); % computing and storing states lastX = A*lastX + Baos*aos_vect(:,k); end end function [x,y,lastX] = doSimulateComplete(lastX, Nsamples, A, Baos, Coutputs, Cstates, Daos, Bnoise, Dnoise, Bcst, Dcst, aos_vect, doTerminate, terminationCond, displayTime, timestep) disp('Running simulate complete...'); %% converting to sparse matrices if numel(A)>0; if(sum(sum(A==0))/numel(A))>0.5; A = sparse(A); end, end if numel(Baos)>0; if(sum(sum(Baos==0))/numel(Baos))>0.5; Baos = sparse(Baos); end, end if numel(Coutputs)>0; if(sum(sum(Coutputs==0))/numel(Coutputs))>0.5; Coutputs = sparse(Coutputs); end, end if numel(Cstates)>0; if(sum(sum(Cstates==0))/numel(Cstates))>0.5; Cstates = sparse(Cstates); end, end if numel(Daos)>0; if(sum(sum(Daos==0))/numel(Daos))>0.5; Daos = sparse(Daos); end, end if numel(Bnoise)>0; if(sum(sum(Bnoise==0))/numel(Bnoise))>0.5; Bnoise = sparse(Bnoise); end, end if numel(Dnoise)>0; if(sum(sum(Dnoise==0))/numel(Dnoise))>0.5; Dnoise = sparse(Dnoise); end, end if numel(Bcst)>0; if(sum(sum(Bcst==0))/numel(Bcst))>0.5; Bcst = sparse(Bcst); end, end if numel(Dcst)>0; if(sum(sum(Dcst==0))/numel(Dcst))>0.5; Dcst = sparse(Dcst); end, end %% initializing fields x = zeros(size(Cstates,1), Nsamples); y = zeros(size(Coutputs,1), Nsamples); Nnoise = size(Bnoise,2); BLOCK_SIZE = min( [ floor(1e6/(size(Baos,2) + size(Baos,1) + size(Bnoise,2) + 1)) , Nsamples]); noise_array = randn(Nnoise, BLOCK_SIZE); lastX = A*lastX + Bcst + Bnoise*noise_array(:,1) + Baos*aos_vect(:,1); % time for displaying remaining time time1=time; %% simulation loop for k = 1:Nsamples %% writing white noise and displaying time if rem(k,BLOCK_SIZE)==0 noise_array = randn(Nnoise, BLOCK_SIZE); if displayTime display( [' simulation time : ',num2str(k*timestep) ]); time2 = time; tloop = floor(time2.utc_epoch_milli-time1.utc_epoch_milli)/1000; display( ['remaining computing time : ',num2str(tloop*(Nsamples-k+1)/k), 's' ]); end end %% evaluating differences knoise = rem(k,BLOCK_SIZE-1)+1; %% computing and storing outputs y(:,k) = Coutputs*lastX + Dcst + Dnoise*noise_array(:,knoise) + Daos*aos_vect(:,k); %% computing and storing states x(:,k) = Cstates*lastX; lastX = A*lastX + Bcst + Bnoise*noise_array(:,knoise) + Baos*aos_vect(:,k); %% checking possible termination condition if doTerminate if eval(terminationCond) x = x(:,1:k); y = y(:,1:k); break; end end end end