├── 1D Simple Harmonic Oscillator
├── README.md
├── SHO_Problem.mlx
├── SMCsampler.m
├── TMCMCsampler.m
├── box_function.m
├── desktop.ini
├── log_likelihood.m
├── model.m
└── proposal_rnd.m
├── 1D Static Spring-Mass System
├── Linear_Problem.mlx
├── README.md
├── SMCsampler.m
├── TMCMCsampler.m
├── desktop.ini
├── log_likelihood.m
└── model.m
├── 2D Eigen-value Problem
├── Eigenvalue_Problem.mlx
├── README.md
├── SMCsampler.m
├── TMCMCsampler.m
├── desktop.ini
├── log_likelihood.m
└── model.m
├── Alternative_TMCMC_Transition_Criteria
├── DenHartogHarmonic.m
├── TMCMCsampler.m
├── TMCMCsampler2.m
├── angleCalc.m
├── areaMe.m
├── blackbox_model.m
├── example_SDOF_System_Coulomb_Friction_numerical.m
├── loglikelihood.m
└── test
├── LICENSE
└── README.md
/1D Simple Harmonic Oscillator/README.md:
--------------------------------------------------------------------------------
1 | ## Instructions:
2 |
3 | * Run the tutorial MATLAB LIVE script: "SHO_Problem.mlx"
4 |
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/SHO_Problem.mlx:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Adolphus8/Bayesian-Model-Updating-Tutorials/f88cbbc1bac5e4caab6f5fd620d76a25c50defa5/1D Simple Harmonic Oscillator/SHO_Problem.mlx
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/SMCsampler.m:
--------------------------------------------------------------------------------
1 | function [output] = SMCsampler(varargin)
2 | %% Sequential Monte Carlo Dynamical (SMC) sampler
3 | %
4 | % This program implements the original Sequential Monte Carlo (SMC) sampling
5 | % class (see paper by N. Chopin (2002): A sequential particle filter method
6 | % for static models - https://www.jstor.org/stable/3879283) and employs
7 | % the use of the Affine-invariant Ensemble Sampler (AIES) proposed by
8 | % Goodman and Weare (2010) to update the samples at each iteration.
9 | %
10 | % This sampler function can be employed in Sequential Bayesian Model
11 | % Updating problems involving:
12 | % - Estimating time-invariant parameter(s) via Online Bayesian Model Updating;
13 | % - Estimating time-varying parameter(s), following a recursive dynamic model;
14 | % - Predicting the time-varying parameter(s) for the next time-step given
15 | % data/observations up to the previous time-step.
16 | %
17 | %--------------------------------------------------------------------------
18 | % Author:
19 | % Adolphus Lye - adolphus.lye@liverpool.ac.uk
20 | %--------------------------------------------------------------------------
21 |
22 | % Parse the information in the name/value pairs:
23 | pnames = {'nsamples','loglikelihoods','dynamic_model',...
24 | 'priorpdf','priorrnd','burnin','lastburnin','thinchain'};
25 |
26 | % Define default values:
27 | dflts = {[], [], @(x) x, [], [], 0, 0, 3};
28 |
29 | [nsamples,loglikelihoods,dynamic_model,priorpdf,prior_rnd,...
30 | burnin,lastBurnin,thinchain] = internal.stats.parseArgs(pnames, dflts, varargin{:});
31 |
32 | %--------------------------------------------------------------------------
33 | %
34 | % Inputs:
35 | % nsamples: Scalar value of the number of samples to be generated from the Posterior;
36 | % loglikelihoods: A M x 1 cell vector of likelihood functions containing the measurements at M different time-steps;
37 | % dynamic_model: A function-handle that relates theta(t+1) and theta(t), where t is the time-step. Output is N x dim;
38 | % priorpdf: Function-handle of the Prior PDF;
39 | % prior_rnd: Function-handle of the Prior random number generator;
40 | % burnin: Number of burn-in for all iterations up to M-1;
41 | % lastBurnin: Number of burn-in for the last iteration;
42 | % stepsize: The stepsize for the Ensemble sampler in the updating step (this is the tuning parameter);
43 | % thinchain: Thin all the chains of the Ensemble sampler by only storing every k'th step (default=3);
44 | %
45 | % Outputs:
46 | % output.samples: A N x dim matrix of Posterior samples;
47 | % output.allsamples: A N x dim x (M+1) array of samples from all iterations;
48 | % output.acceptance: A M x 1 vector of acceptance rates for all iterations;
49 | % output.log_evidence: A (M+1) x 1 vector of the logarithmic of the evidence;
50 | % output.step: A M x 1 vector of step-size;
51 | % output.indicator: A M x 1 vector of indicators denoting if resampling
52 | % has occured for any iterations (1 = Yes, 0 = No);
53 | %
54 | %--------------------------------------------------------------------------
55 |
56 | %% Number of cores
57 | if ~isempty(gcp('nocreate'))
58 | pool = gcp;
59 | Ncores = pool.NumWorkers;
60 | fprintf('SMC is running on %d cores.\n', Ncores);
61 | end
62 |
63 | %% Initialize: Obtain N samples from the Prior PDF
64 |
65 | fprintf('Start SMC procedure ... \n');
66 |
67 | prior_initial = priorpdf; % Define initial Prior PDF
68 | thetaj = prior_rnd(nsamples); % theta0 = N x dim
69 | Dimensions = size(thetaj, 2); % Dimensionality of theta, dim
70 |
71 | % Initialization of matrices and vectors:
72 | thetaj1 = zeros(nsamples, Dimensions);
73 | log_evidence = zeros(size(loglikelihoods,1)+1,1); % Initiate empty vector for log evidence
74 | log_evidence(1) = 0;
75 |
76 | acceptance = zeros(size(loglikelihoods,1),1);
77 |
78 | % Samples from filter distribution, P(theta(t)|Data(1:t)):
79 | allsamples = zeros(size(thetaj,1), size(thetaj,2), size(loglikelihoods,1)+1);
80 | allsamples(:,:,1) = thetaj;
81 |
82 | % Statistics from predictive distribution, P(theta(t+1)|Data(1:t)):
83 | predictive_samples = zeros(size(thetaj,1), size(thetaj,2), size(loglikelihoods,1));
84 |
85 | % Resampling indicator vector:
86 | indicator = zeros(length(loglikelihoods), 1);
87 | % Note: This indicator vector returns a 1 for the iteration(s) where
88 | % resampling is initiated and 0 otherwise.
89 |
90 | %% Main sampling loop
91 | for iter = 1:length(loglikelihoods)
92 |
93 | fprintf('SMC: Iteration j = %2d \n', iter);
94 |
95 | loglikelihood = loglikelihoods{iter};
96 |
97 | % Compute loglikelihood values for each sample:
98 | logL = zeros(nsamples,1);
99 | for l = 1:nsamples
100 | logL(l) = loglikelihood(thetaj(l,:));
101 | end
102 |
103 | % Error check:
104 | if any(isinf(logL))
105 | error('The prior distribution is too far from the true region');
106 | end
107 |
108 | %% Compute weights of the samples, wj:
109 |
110 | % To compute the nominal weights:
111 | fprintf('Computing the weights ...\n');
112 | wj = exp(logL);
113 |
114 | % To compute the log evidence for the current iteration:
115 | log_evidence(iter+1) = log(mean(wj)) + log_evidence(iter);
116 |
117 | % Check step for wj:
118 | for i = 1:nsamples
119 | if wj(i) == 0
120 | wj(i) = 1e-100;
121 | end
122 | end
123 |
124 | wj_norm = wj./sum(wj); % To normalise the weights
125 |
126 | %% Check step - Compute the sum of wj_norm and see if it is < nsamples/2:
127 |
128 | fprintf('Computing effective sample size ... \n');
129 | Neff = 1/(sum(wj_norm.^2));
130 | threshold = nsamples/2;
131 |
132 | %% Resampling step (conditional if Neff < threshold):
133 |
134 | if Neff < threshold
135 | fprintf('Resampling step initiated ... \n');
136 |
137 | dx = randsample(nsamples, nsamples, true, wj_norm);
138 |
139 | thetaj_resampled = zeros(nsamples, Dimensions);
140 | for d = 1:nsamples
141 | thetaj_resampled(d,:) = thetaj(dx(d),:);
142 | end
143 |
144 | thetaj = thetaj_resampled;
145 | wj_norm = (1/nsamples).*ones(nsamples,1);
146 | indicator(iter) = 1;
147 |
148 | end
149 |
150 | %% Update the samples according to the current Posterior using MH sampler:
151 |
152 | % Define the logposterior:
153 | log_posterior = @(x) log(priorpdf(x)) + loglikelihood(x);
154 |
155 | % Weighted mean for Proposal distribution
156 | mu = zeros(1, Dimensions);
157 | for l = 1:nsamples
158 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
159 | end
160 |
161 | % Covariance matrix for Proposal distribution:
162 | cov_gauss = zeros(Dimensions);
163 | for k = 1:nsamples
164 | tk_mu = thetaj(k,:) - mu;
165 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
166 | end
167 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
168 |
169 | % Define the Proposal distribution:
170 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, prior_initial); % q(x,y) = q(x|y).
171 | proprnd = @(x) prop_rnd(x, cov_gauss, prior_initial);
172 |
173 | if iter == length(loglikelihoods)
174 | burnin = lastBurnin;
175 | end
176 |
177 | %% Start N different Markov chains
178 | fprintf('Markov chains ...\n\n');
179 |
180 | idx = randsample(nsamples, nsamples, true, wj_norm);
181 | for i = 1:nsamples % For parallel, type: parfor
182 | [thetaj1(i,:), acceptance_rate(i)] = mhsample(thetaj(idx(i),:), 1, ...
183 | 'logpdf', log_posterior, ...
184 | 'proppdf', proppdf, ...
185 | 'proprnd', proprnd, ...
186 | 'thin', thinchain, ...
187 | 'burnin', burnin);
188 | end
189 | fprintf('\n');
190 | acceptance(iter) = mean(acceptance_rate); % To store the acceptance rate values
191 |
192 | %% Prediction step:
193 |
194 | % Define the Predictive distribution of the samples, P(theta(t+1)|Data(t)):
195 | predictive_samples(:,:,iter) = dynamic_model(thetaj1);
196 |
197 | % Compute the Bandwidth vector for the kernel density function:
198 | pred_samps = predictive_samples(:,:,iter);
199 |
200 | bw = zeros(Dimensions,1);
201 | for dim = 1:Dimensions
202 | bw(dim) = std(pred_samps(:,dim)) .* (4/((Dimensions + 2) .* nsamples)).^(1/(Dimensions + 4));
203 | end
204 |
205 | % Define the Predictive PDF, P(theta(t+1)|Data(t)) using mvksdensity:
206 | pred_pdf = @(x) mvksdensity(pred_samps, x, 'Bandwidth', bw);
207 |
208 | %% Prepare for the next iteration:
209 |
210 | allsamples(:,:,iter+1) = thetaj1;
211 | thetaj = pred_samps;
212 | priorpdf = @(x) pred_pdf(x);
213 |
214 | end
215 |
216 | %% Description of outputs:
217 |
218 | output.samples = thetaj; % To only show samples from the final filter distribution
219 | output.allsamples = allsamples; % To only show all filter samples across all iterations
220 | output.prediction = predictive_samples; % To only show all prediction samples across all iterations
221 | output.acceptance = acceptance; % To show the mean acceptance rates for all iterations
222 | output.log_evidence = log_evidence; % To show the (M+1) x 1 vector of the logarithmic of the evidence;
223 | output.indicator = indicator; % To indicate the iterations whereby resampling took place (denoted by 1s).
224 |
225 | fprintf('End of SMC procedure. \n\n');
226 |
227 | return; % End
228 |
229 | function proppdf = prop_pdf(x, mu, covmat, box)
230 | % This is the Proposal PDF for the Markov Chain.
231 |
232 | % Box function is the Prior PDF in the feasible region.
233 | % So if a point is out of bounds, this function will
234 | % return 0.
235 |
236 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
237 |
238 | return;
239 |
240 |
241 | function proprnd = prop_rnd(mu, covmat, box)
242 | % Sampling from the proposal PDF for the Markov Chain.
243 |
244 | while true
245 | proprnd = mvnrnd(mu, covmat, 1);
246 | if box(proprnd)
247 | % The box function is the Prior PDF in the feasible region.
248 | % If a point is out of bounds, this function will return 0 = false.
249 | break;
250 | end
251 | end
252 |
253 | return
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/TMCMCsampler.m:
--------------------------------------------------------------------------------
1 | function [samples_fT_D, log_fD] = TMCMCsampler(varargin)
2 | %% Transitional Markov Chain Monte Carlo sampler
3 | %
4 | % This program implements a method described in:
5 | % Ching, J. and Chen, Y. (2007). "Transitional Markov Chain Monte Carlo
6 | % Method for Bayesian Model Updating, Model Class Selection, and Model
7 | % Averaging." J. Eng. Mech., 133(7), 816-832.
8 | %
9 | % Usage:
10 | % [samples_fT_D, fD] = tmcmc_v1(fD_T, fT, sample_from_fT, N);
11 | %
12 | % where:
13 | %
14 | % inputs:
15 | % log_fD_T = function handle of log(fD_T(t)), Loglikelihood
16 | %
17 | % fT = function handle of fT(t), Prior PDF
18 | %
19 | % sample_from_fT = handle to a function that samples from of fT(t),
20 | % Sampling rule function from Prior PDF
21 | %
22 | % nsamples = number of samples of fT_D, Posterior, to generate
23 | %
24 | % outputs:
25 | % samples_fT_D = samples of fT_D (N x D) = samples from Posterior
26 | % distribution
27 | %
28 | % log_fD = log(evidence) = log(normalization constant)
29 |
30 | % ------------------------------------------------------------------------
31 | % who when observations
32 | %--------------------------------------------------------------------------
33 | % Diego Andres Alvarez Jul-24-2013 First algorithm
34 | %--------------------------------------------------------------------------
35 | % Diego Andres Alvarez - daalvarez@unal.edu.co
36 | % Edoardo Patelli - edoardo.patelli@strath.ac.uk
37 |
38 | % parse the information in the name/value pairs:
39 | pnames = {'nsamples','loglikelihood','priorpdf','priorrnd','burnin','lastburnin','beta'};
40 |
41 | dflts = {[],[],[],[],[],0,0.2}; % define default values
42 |
43 | [nsamples,loglikelihood,priorpdf,prior_rnd,burnin,lastBurnin,beta] = ...
44 | internal.stats.parseArgs(pnames, dflts, varargin{:});
45 |
46 | %% Obtain N samples from the prior pdf f(T)
47 | j = 0; % Initialise loop for the transitional likelihood
48 | thetaj = prior_rnd(nsamples); % theta0 = N x D
49 | pj = 0; % p0 = 0 (initial tempering parameter)
50 | Dimensions = size(thetaj, 2); % size of the vector theta
51 |
52 | %% Initialization of matrices and vectors
53 | thetaj1 = zeros(nsamples, Dimensions);
54 | %log_fD_T_thetaj = zeros(nsamples,1);
55 |
56 | %% Main loop
57 | while pj < 1
58 | j = j+1;
59 |
60 | %% Calculate the tempering parameter p(j+1):
61 | for l = 1:nsamples
62 | log_fD_T_thetaj(l) = loglikelihood(thetaj(l,:));
63 | end
64 | if any(isinf(log_fD_T_thetaj))
65 | error('The prior distribution is too far from the true region');
66 | end
67 | pj1 = calculate_pj1(log_fD_T_thetaj, pj);
68 | fprintf('TMCMC: Iteration j = %2d, pj1 = %f\n', j, pj1);
69 |
70 | %% Compute the plausibility weight for each sample wrt f_{j+1}
71 | fprintf('Computing the weights ...\n');
72 | % wj = fD_T(thetaj).^(pj1-pj); % N x 1 (eq 12)
73 | a = (pj1-pj)*log_fD_T_thetaj;
74 | wj = exp(a);
75 | wj_norm = wj./sum(wj); % normalization of the weights
76 |
77 | %% Compute S(j) = E[w{j}] (eq 15)
78 | S(j) = mean(wj);
79 |
80 | %% Do the resampling step to obtain N samples from f_{j+1}(theta) and
81 | % then perform Metropolis-Hastings on each of these samples using as a
82 | % stationary PDF "fj1"
83 | % fj1 = @(t) fT(t).*log_fD_T(t).^pj1; % stationary PDF (eq 11) f_{j+1}(theta)
84 | log_posterior = @(t) log(priorpdf(t)) + pj1*loglikelihood(t);
85 |
86 |
87 | % and using as proposal PDF a Gaussian centered at thetaj(idx,:) and
88 | % with covariance matrix equal to an scaled version of the covariance
89 | % matrix of fj1:
90 |
91 | % weighted mean
92 | mu = zeros(1, Dimensions);
93 | for l = 1:nsamples
94 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
95 | end
96 |
97 | % scaled covariance matrix of fj1 (eq 17)
98 | cov_gauss = zeros(Dimensions);
99 | for k = 1:nsamples
100 | % this formula is slightly different to eq 17 (the transpose)
101 | % because of the size of the vectors)m and because Ching and Chen
102 | % forgot to normalize the weight wj:
103 | tk_mu = thetaj(k,:) - mu;
104 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
105 | end
106 | cov_gauss = beta^2 * cov_gauss;
107 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
108 |
109 | % Define the Proposal distribution:
110 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, priorpdf); %q(x,y) = q(x|y).
111 | proprnd = @(x) prop_rnd(x, cov_gauss, priorpdf);
112 |
113 | %% During the last iteration we require to do a better burnin in order
114 | % to guarantee the quality of the samples:
115 | if pj1 == 1
116 | burnin = lastBurnin;
117 | end;
118 |
119 | %% Start N different Markov chains
120 | fprintf('Markov chains ...\n\n');
121 | idx = randsample(nsamples, nsamples, true, wj_norm);
122 | for i = 1:nsamples % For parallel, type: parfor
123 | %% Sample one point with probability wj_norm
124 |
125 | % smpl = mhsample(start, nsamples,
126 | % 'pdf', pdf, 'proppdf', proppdf, 'proprnd', proprnd);
127 | % start = row vector containing the start value of the Markov Chain,
128 | % nsamples = number of samples to be generated
129 | [thetaj1(i,:), acceptance_rate] = mhsample(thetaj(idx(i), :), 1, ...
130 | 'logpdf', log_posterior, ...
131 | 'proppdf', proppdf, ...
132 | 'proprnd', proprnd, ...
133 | 'thin', 3, ...
134 | 'burnin', burnin);
135 | % According to Cheung and Beck (2009) - Bayesian model updating ...,
136 | % the initial samples from reweighting and the resample of samples of
137 | % fj, in general, do not exactly follow fj1, so that the Markov
138 | % chains must "burn-in" before samples follow fj1, requiring a large
139 | % amount of samples to be generated for each level.
140 |
141 | %% Adjust the acceptance rate (optimal = 23%)
142 | % See: http://www.dms.umontreal.ca/~bedard/Beyond_234.pdf
143 | %{
144 | if acceptance_rate < 0.3
145 | % Many rejections means an inefficient chain (wasted computation
146 | %time), decrease the variance
147 | beta = 0.99*beta;
148 | elseif acceptance_rate > 0.5
149 | % High acceptance rate: Proposed jumps are very close to current
150 | % location, increase the variance
151 | beta = 1.01*beta;
152 | end
153 | %}
154 | end
155 | fprintf('\n');
156 |
157 | %% Prepare for the next iteration
158 | thetaj = thetaj1;
159 | pj = pj1;
160 | end
161 |
162 | % TMCMC provides N samples distributed according to the Posterior distribution, f(T|D)
163 | samples_fT_D = thetaj;
164 |
165 | % estimation of f(D) -- this is the normalization constant in Bayes
166 | log_fD = sum(log(S(1:j)));
167 |
168 | return; % End
169 |
170 |
171 | %% Calculate the tempering parameter p(j+1)
172 | function pj1 = calculate_pj1(log_fD_T_thetaj, pj)
173 | % find pj1 such that COV <= threshold, that is
174 | %
175 | % std(wj)
176 | % --------- <= threshold
177 | % mean(wj)
178 | %
179 | % here
180 | % size(thetaj) = N x D,
181 | % wj = fD_T(thetaj).^(pj1 - pj)
182 | % e = pj1 - pj
183 |
184 | threshold = 1; % 100% = threshold on the COV
185 |
186 | % wj = @(e) fD_T_thetaj^e; % N x 1
187 | % Note the following trick in order to calculate e:
188 | % Take into account that e>=0
189 | wj = @(e) exp(abs(e)*log_fD_T_thetaj); % N x 1
190 | %fmin = @(e) std(wj(e))/mean(wj(e)) - threshold;
191 | fmin = @(e) std(wj(e)) - threshold*mean(wj(e)) + realmin;
192 | e = abs(fzero(fmin, 0)); % e is >= 0, and fmin is an even function
193 | if isnan(e)
194 | error('There is an error finding e');
195 | end
196 |
197 | pj1 = min(1, pj + e);
198 |
199 | return; % End
200 |
201 | function proppdf = prop_pdf(x, mu, covmat, box)
202 | % This is the Proposal PDF for the Markov Chain.
203 |
204 | % Box function is the Prior PDF in the feasible region.
205 | % So if a point is out of bounds, this function will
206 | % return 0.
207 |
208 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
209 |
210 | return;
211 |
212 |
213 | function proprnd = prop_rnd(mu, covmat, box)
214 | % Sampling from the proposal PDF for the Markov Chain.
215 |
216 | while true
217 | proprnd = mvnrnd(mu, covmat, 1);
218 | if box(proprnd)
219 | % The box function is the Prior PDF in the feasible region.
220 | % If a point is out of bounds, this function will return 0 = false.
221 | break;
222 | end
223 | end
224 |
225 | return
226 |
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/box_function.m:
--------------------------------------------------------------------------------
1 | function [BoxFunction] = box_function(theta, parameters)
2 |
3 | % The Box Function serves as an indicator function such that should the
4 | % Candidate samples fall outside the range of values of the Uniform Prior,
5 | % the function returns a 0 and returns a 1 if the values of the Candidate
6 | % samples fall within the range of values of the Uniform Prior.
7 |
8 |
9 | if theta < parameters(1) || theta > parameters(2)
10 |
11 | BoxFunction = 0;
12 |
13 | else
14 |
15 | BoxFunction = 1;
16 |
17 | end
18 |
19 | end
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/desktop.ini:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Adolphus8/Bayesian-Model-Updating-Tutorials/f88cbbc1bac5e4caab6f5fd620d76a25c50defa5/1D Simple Harmonic Oscillator/desktop.ini
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/log_likelihood.m:
--------------------------------------------------------------------------------
1 | function logl = log_likelihood(stiffness, modelInput, measurements, standard_deviation, ModelHandle)
2 | % Calculation of the log_likelihood for 1D Simple Harmonic Oscillator:
3 | %
4 | % USAGE:
5 | % logl = log_likelihood(stiffness, mass, measurements, standard_deviation, ModelHandle)
6 | %
7 | % INPUTS:
8 | % stiffness = epistemic parameter k [Nsamples x 1]
9 | % mass = model input [Nobservations x 1]
10 | % measurements = experimental observations [Nobservations x 1]
11 | % standard_deviation = the standard deviation of the log likelihood function [scalar]
12 | % ModelHandle = the function handle of the model (see file "model.m")
13 | %
14 | % OUTPUTS:
15 | % logl = loglikelihood function for the set of estimated stiffness values k and
16 | % the measurements
17 | % .
18 | % log1 = 1 x 1
19 | %
20 |
21 | %% Evaluate the model:
22 | nchains=size(stiffness,1);
23 | %ndims=size(stiffness,2);
24 | logl=zeros(nchains,1);
25 |
26 | for n=1:nchains
27 | modelOutput = ModelHandle(stiffness(n,:),modelInput);
28 |
29 | % Note: Details to the model can be found in the file: "model.m"
30 |
31 | %% Compute the log-likelihood:
32 |
33 | logl(n) = -0.5 * (1/standard_deviation)^2 *(measurements - modelOutput)' * (measurements - modelOutput);
34 |
35 | end
36 |
37 |
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/model.m:
--------------------------------------------------------------------------------
1 | function [frequency] = model(stiffness,mass)
2 |
3 | %MODEL: 1-Degree of freedom simple harmonic oscillator system
4 |
5 | frequency = sqrt(stiffness./mass);
6 |
7 | end
--------------------------------------------------------------------------------
/1D Simple Harmonic Oscillator/proposal_rnd.m:
--------------------------------------------------------------------------------
1 | function [proprnd] = proposal_rnd(CurrentSample, Tuning_mcmc, NumberOfChains, BoxFunction)
2 |
3 | % This is a modified Proposal random number generator from the Normal
4 | % Proposal distribution. The normrnd function is now multiplied by the Box
5 | % function such that if the generated proposal sample values fall outide
6 | % the range of values as stipulated by the Prior distribution, the function
7 | % returns a 0 immediately and the sample value is rejected. This serves to
8 | % prevent any values which fall outside the range of Prior values from
9 | % being accepted.
10 |
11 | proprnd_nominal = normrnd(CurrentSample,Tuning_mcmc,NumberOfChains,1);
12 |
13 | % Initiate the storing of the array of Proposal sample values with an empty array:
14 | proprnd = zeros(NumberOfChains,1);
15 |
16 | % To store each array value of proprnd_nominal:
17 | for i = 1:NumberOfChains
18 | proprnd(i) = proprnd_nominal(i) .* BoxFunction(proprnd_nominal(i));
19 | end
20 |
21 | end
--------------------------------------------------------------------------------
/1D Static Spring-Mass System/Linear_Problem.mlx:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Adolphus8/Bayesian-Model-Updating-Tutorials/f88cbbc1bac5e4caab6f5fd620d76a25c50defa5/1D Static Spring-Mass System/Linear_Problem.mlx
--------------------------------------------------------------------------------
/1D Static Spring-Mass System/README.md:
--------------------------------------------------------------------------------
1 | ## Instructions:
2 |
3 | * Run the tutorial MATLAB LIVE script: "Linear_Problem.mlx"
4 |
--------------------------------------------------------------------------------
/1D Static Spring-Mass System/SMCsampler.m:
--------------------------------------------------------------------------------
1 | function [output] = SMCsampler(varargin)
2 | %% Sequential Monte Carlo Dynamical (SMC) sampler
3 | %
4 | % This program implements the original Sequential Monte Carlo (SMC) sampling
5 | % class (see paper by N. Chopin (2002): A sequential particle filter method
6 | % for static models - https://www.jstor.org/stable/3879283) and employs
7 | % the use of the Affine-invariant Ensemble Sampler (AIES) proposed by
8 | % Goodman and Weare (2010) to update the samples at each iteration.
9 | %
10 | % This sampler function can be employed in Sequential Bayesian Model
11 | % Updating problems involving:
12 | % - Estimating time-invariant parameter(s) via Online Bayesian Model Updating;
13 | % - Estimating time-varying parameter(s), following a recursive dynamic model;
14 | % - Predicting the time-varying parameter(s) for the next time-step given
15 | % data/observations up to the previous time-step.
16 | %
17 | %--------------------------------------------------------------------------
18 | % Author:
19 | % Adolphus Lye - adolphus.lye@liverpool.ac.uk
20 | %--------------------------------------------------------------------------
21 |
22 | % Parse the information in the name/value pairs:
23 | pnames = {'nsamples','loglikelihoods','dynamic_model',...
24 | 'priorpdf','priorrnd','burnin','lastburnin','thinchain'};
25 |
26 | % Define default values:
27 | dflts = {[], [], @(x) x, [], [], 0, 0, 3};
28 |
29 | [nsamples,loglikelihoods,dynamic_model,priorpdf,prior_rnd,...
30 | burnin,lastBurnin,thinchain] = internal.stats.parseArgs(pnames, dflts, varargin{:});
31 |
32 | %--------------------------------------------------------------------------
33 | %
34 | % Inputs:
35 | % nsamples: Scalar value of the number of samples to be generated from the Posterior;
36 | % loglikelihoods: A M x 1 cell vector of likelihood functions containing the measurements at M different time-steps;
37 | % dynamic_model: A function-handle that relates theta(t+1) and theta(t), where t is the time-step. Output is N x dim;
38 | % priorpdf: Function-handle of the Prior PDF;
39 | % prior_rnd: Function-handle of the Prior random number generator;
40 | % burnin: Number of burn-in for all iterations up to M-1;
41 | % lastBurnin: Number of burn-in for the last iteration;
42 | % stepsize: The stepsize for the Ensemble sampler in the updating step (this is the tuning parameter);
43 | % thinchain: Thin all the chains of the Ensemble sampler by only storing every k'th step (default=3);
44 | %
45 | % Outputs:
46 | % output.samples: A N x dim matrix of Posterior samples;
47 | % output.allsamples: A N x dim x (M+1) array of samples from all iterations;
48 | % output.acceptance: A M x 1 vector of acceptance rates for all iterations;
49 | % output.log_evidence: A (M+1) x 1 vector of the logarithmic of the evidence;
50 | % output.step: A M x 1 vector of step-size;
51 | % output.indicator: A M x 1 vector of indicators denoting if resampling
52 | % has occured for any iterations (1 = Yes, 0 = No);
53 | %
54 | %--------------------------------------------------------------------------
55 |
56 | %% Number of cores
57 | if ~isempty(gcp('nocreate'))
58 | pool = gcp;
59 | Ncores = pool.NumWorkers;
60 | fprintf('SMC is running on %d cores.\n', Ncores);
61 | end
62 |
63 | %% Initialize: Obtain N samples from the Prior PDF
64 |
65 | fprintf('Start SMC procedure ... \n');
66 |
67 | prior_initial = priorpdf; % Define initial Prior PDF
68 | thetaj = prior_rnd(nsamples); % theta0 = N x dim
69 | Dimensions = size(thetaj, 2); % Dimensionality of theta, dim
70 |
71 | % Initialization of matrices and vectors:
72 | thetaj1 = zeros(nsamples, Dimensions);
73 | log_evidence = zeros(size(loglikelihoods,1)+1,1); % Initiate empty vector for log evidence
74 | log_evidence(1) = 0;
75 |
76 | acceptance = zeros(size(loglikelihoods,1),1);
77 |
78 | % Samples from filter distribution, P(theta(t)|Data(1:t)):
79 | allsamples = zeros(size(thetaj,1), size(thetaj,2), size(loglikelihoods,1)+1);
80 | allsamples(:,:,1) = thetaj;
81 |
82 | % Statistics from predictive distribution, P(theta(t+1)|Data(1:t)):
83 | predictive_samples = zeros(size(thetaj,1), size(thetaj,2), size(loglikelihoods,1));
84 |
85 | % Resampling indicator vector:
86 | indicator = zeros(length(loglikelihoods), 1);
87 | % Note: This indicator vector returns a 1 for the iteration(s) where
88 | % resampling is initiated and 0 otherwise.
89 |
90 | %% Main sampling loop
91 | for iter = 1:length(loglikelihoods)
92 |
93 | fprintf('SMC: Iteration j = %2d \n', iter);
94 |
95 | loglikelihood = loglikelihoods{iter};
96 |
97 | % Compute loglikelihood values for each sample:
98 | logL = zeros(nsamples,1);
99 | for l = 1:nsamples
100 | logL(l) = loglikelihood(thetaj(l,:));
101 | end
102 |
103 | % Error check:
104 | if any(isinf(logL))
105 | error('The prior distribution is too far from the true region');
106 | end
107 |
108 | %% Compute weights of the samples, wj:
109 |
110 | % To compute the nominal weights:
111 | fprintf('Computing the weights ...\n');
112 | wj = exp(logL);
113 |
114 | % To compute the log evidence for the current iteration:
115 | log_evidence(iter+1) = log(mean(wj)) + log_evidence(iter);
116 |
117 | % Check step for wj:
118 | for i = 1:nsamples
119 | if wj(i) == 0
120 | wj(i) = 1e-100;
121 | end
122 | end
123 |
124 | wj_norm = wj./sum(wj); % To normalise the weights
125 |
126 | %% Check step - Compute the sum of wj_norm and see if it is < nsamples/2:
127 |
128 | fprintf('Computing effective sample size ... \n');
129 | Neff = 1/(sum(wj_norm.^2));
130 | threshold = nsamples/2;
131 |
132 | %% Resampling step (conditional if Neff < threshold):
133 |
134 | if Neff < threshold
135 | fprintf('Resampling step initiated ... \n');
136 |
137 | dx = randsample(nsamples, nsamples, true, wj_norm);
138 |
139 | thetaj_resampled = zeros(nsamples, Dimensions);
140 | for d = 1:nsamples
141 | thetaj_resampled(d,:) = thetaj(dx(d),:);
142 | end
143 |
144 | thetaj = thetaj_resampled;
145 | wj_norm = (1/nsamples).*ones(nsamples,1);
146 | indicator(iter) = 1;
147 |
148 | end
149 |
150 | %% Update the samples according to the current Posterior using MH sampler:
151 |
152 | % Define the logposterior:
153 | log_posterior = @(x) log(priorpdf(x)) + loglikelihood(x);
154 |
155 | % Weighted mean for Proposal distribution
156 | mu = zeros(1, Dimensions);
157 | for l = 1:nsamples
158 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
159 | end
160 |
161 | % Covariance matrix for Proposal distribution:
162 | cov_gauss = zeros(Dimensions);
163 | for k = 1:nsamples
164 | tk_mu = thetaj(k,:) - mu;
165 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
166 | end
167 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
168 |
169 | % Define the Proposal distribution:
170 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, prior_initial); % q(x,y) = q(x|y).
171 | proprnd = @(x) prop_rnd(x, cov_gauss, prior_initial);
172 |
173 | if iter == length(loglikelihoods)
174 | burnin = lastBurnin;
175 | end
176 |
177 | %% Start N different Markov chains
178 | fprintf('Markov chains ...\n\n');
179 |
180 | idx = randsample(nsamples, nsamples, true, wj_norm);
181 | for i = 1:nsamples % For parallel, type: parfor
182 | [thetaj1(i,:), acceptance_rate(i)] = mhsample(thetaj(idx(i),:), 1, ...
183 | 'logpdf', log_posterior, ...
184 | 'proppdf', proppdf, ...
185 | 'proprnd', proprnd, ...
186 | 'thin', thinchain, ...
187 | 'burnin', burnin);
188 | end
189 | fprintf('\n');
190 | acceptance(iter) = mean(acceptance_rate); % To store the acceptance rate values
191 |
192 | %% Prediction step:
193 |
194 | % Define the Predictive distribution of the samples, P(theta(t+1)|Data(t)):
195 | predictive_samples(:,:,iter) = dynamic_model(thetaj1);
196 |
197 | % Compute the Bandwidth vector for the kernel density function:
198 | pred_samps = predictive_samples(:,:,iter);
199 |
200 | bw = zeros(Dimensions,1);
201 | for dim = 1:Dimensions
202 | bw(dim) = std(pred_samps(:,dim)) .* (4/((Dimensions + 2) .* nsamples)).^(1/(Dimensions + 4));
203 | end
204 |
205 | % Define the Predictive PDF, P(theta(t+1)|Data(t)) using mvksdensity:
206 | pred_pdf = @(x) mvksdensity(pred_samps, x, 'Bandwidth', bw);
207 |
208 | %% Prepare for the next iteration:
209 |
210 | allsamples(:,:,iter+1) = thetaj1;
211 | thetaj = pred_samps;
212 | priorpdf = @(x) pred_pdf(x);
213 |
214 | end
215 |
216 | %% Description of outputs:
217 |
218 | output.samples = thetaj; % To only show samples from the final filter distribution
219 | output.allsamples = allsamples; % To only show all filter samples across all iterations
220 | output.prediction = predictive_samples; % To only show all prediction samples across all iterations
221 | output.acceptance = acceptance; % To show the mean acceptance rates for all iterations
222 | output.log_evidence = log_evidence; % To show the (M+1) x 1 vector of the logarithmic of the evidence;
223 | output.indicator = indicator; % To indicate the iterations whereby resampling took place (denoted by 1s).
224 |
225 | fprintf('End of SMC procedure. \n\n');
226 |
227 | return; % End
228 |
229 | function proppdf = prop_pdf(x, mu, covmat, box)
230 | % This is the Proposal PDF for the Markov Chain.
231 |
232 | % Box function is the Prior PDF in the feasible region.
233 | % So if a point is out of bounds, this function will
234 | % return 0.
235 |
236 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
237 |
238 | return;
239 |
240 |
241 | function proprnd = prop_rnd(mu, covmat, box)
242 | % Sampling from the proposal PDF for the Markov Chain.
243 |
244 | while true
245 | proprnd = mvnrnd(mu, covmat, 1);
246 | if box(proprnd)
247 | % The box function is the Prior PDF in the feasible region.
248 | % If a point is out of bounds, this function will return 0 = false.
249 | break;
250 | end
251 | end
252 |
253 | return
--------------------------------------------------------------------------------
/1D Static Spring-Mass System/TMCMCsampler.m:
--------------------------------------------------------------------------------
1 | function [samples_fT_D, log_fD] = TMCMCsampler(varargin)
2 | %% Transitional Markov Chain Monte Carlo sampler
3 | %
4 | % This program implements a method described in:
5 | % Ching, J. and Chen, Y. (2007). "Transitional Markov Chain Monte Carlo
6 | % Method for Bayesian Model Updating, Model Class Selection, and Model
7 | % Averaging." J. Eng. Mech., 133(7), 816-832.
8 | %
9 | % Usage:
10 | % [samples_fT_D, fD] = tmcmc_v1(fD_T, fT, sample_from_fT, N);
11 | %
12 | % where:
13 | %
14 | % inputs:
15 | % log_fD_T = function handle of log(fD_T(t)), Loglikelihood
16 | %
17 | % fT = function handle of fT(t), Prior PDF
18 | %
19 | % sample_from_fT = handle to a function that samples from of fT(t),
20 | % Sampling rule function from Prior PDF
21 | %
22 | % nsamples = number of samples of fT_D, Posterior, to generate
23 | %
24 | % outputs:
25 | % samples_fT_D = samples of fT_D (N x D) = samples from Posterior
26 | % distribution
27 | %
28 | % log_fD = log(evidence) = log(normalization constant)
29 |
30 | % ------------------------------------------------------------------------
31 | % who when observations
32 | %--------------------------------------------------------------------------
33 | % Diego Andres Alvarez Jul-24-2013 First algorithm
34 | %--------------------------------------------------------------------------
35 | % Diego Andres Alvarez - daalvarez@unal.edu.co
36 | % Edoardo Patelli - edoardo.patelli@strath.ac.uk
37 |
38 | % parse the information in the name/value pairs:
39 | pnames = {'nsamples','loglikelihood','priorpdf','priorrnd','burnin','lastburnin','beta'};
40 |
41 | dflts = {[],[],[],[],[],0,0.2}; % define default values
42 |
43 | [nsamples,loglikelihood,priorpdf,prior_rnd,burnin,lastBurnin,beta] = ...
44 | internal.stats.parseArgs(pnames, dflts, varargin{:});
45 |
46 | %% Obtain N samples from the prior pdf f(T)
47 | j = 0; % Initialise loop for the transitional likelihood
48 | thetaj = prior_rnd(nsamples); % theta0 = N x D
49 | pj = 0; % p0 = 0 (initial tempering parameter)
50 | Dimensions = size(thetaj, 2); % size of the vector theta
51 |
52 | %% Initialization of matrices and vectors
53 | thetaj1 = zeros(nsamples, Dimensions);
54 | %log_fD_T_thetaj = zeros(nsamples,1);
55 |
56 | %% Main loop
57 | while pj < 1
58 | j = j+1;
59 |
60 | %% Calculate the tempering parameter p(j+1):
61 | for l = 1:nsamples
62 | log_fD_T_thetaj(l) = loglikelihood(thetaj(l,:));
63 | end
64 | if any(isinf(log_fD_T_thetaj))
65 | error('The prior distribution is too far from the true region');
66 | end
67 | pj1 = calculate_pj1(log_fD_T_thetaj, pj);
68 | fprintf('TMCMC: Iteration j = %2d, pj1 = %f\n', j, pj1);
69 |
70 | %% Compute the plausibility weight for each sample wrt f_{j+1}
71 | fprintf('Computing the weights ...\n');
72 | % wj = fD_T(thetaj).^(pj1-pj); % N x 1 (eq 12)
73 | a = (pj1-pj)*log_fD_T_thetaj;
74 | wj = exp(a);
75 | wj_norm = wj./sum(wj); % normalization of the weights
76 |
77 | %% Compute S(j) = E[w{j}] (eq 15)
78 | S(j) = mean(wj);
79 |
80 | %% Do the resampling step to obtain N samples from f_{j+1}(theta) and
81 | % then perform Metropolis-Hastings on each of these samples using as a
82 | % stationary PDF "fj1"
83 | % fj1 = @(t) fT(t).*log_fD_T(t).^pj1; % stationary PDF (eq 11) f_{j+1}(theta)
84 | log_posterior = @(t) log(priorpdf(t)) + pj1*loglikelihood(t);
85 |
86 |
87 | % and using as proposal PDF a Gaussian centered at thetaj(idx,:) and
88 | % with covariance matrix equal to an scaled version of the covariance
89 | % matrix of fj1:
90 |
91 | % weighted mean
92 | mu = zeros(1, Dimensions);
93 | for l = 1:nsamples
94 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
95 | end
96 |
97 | % scaled covariance matrix of fj1 (eq 17)
98 | cov_gauss = zeros(Dimensions);
99 | for k = 1:nsamples
100 | % this formula is slightly different to eq 17 (the transpose)
101 | % because of the size of the vectors)m and because Ching and Chen
102 | % forgot to normalize the weight wj:
103 | tk_mu = thetaj(k,:) - mu;
104 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
105 | end
106 | cov_gauss = beta^2 * cov_gauss;
107 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
108 |
109 | % Define the Proposal distribution:
110 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, priorpdf); %q(x,y) = q(x|y).
111 | proprnd = @(x) prop_rnd(x, cov_gauss, priorpdf);
112 |
113 | %% During the last iteration we require to do a better burnin in order
114 | % to guarantee the quality of the samples:
115 | if pj1 == 1
116 | burnin = lastBurnin;
117 | end;
118 |
119 | %% Start N different Markov chains
120 | fprintf('Markov chains ...\n\n');
121 | idx = randsample(nsamples, nsamples, true, wj_norm);
122 | for i = 1:nsamples % For parallel, type: parfor
123 | %% Sample one point with probability wj_norm
124 |
125 | % smpl = mhsample(start, nsamples,
126 | % 'pdf', pdf, 'proppdf', proppdf, 'proprnd', proprnd);
127 | % start = row vector containing the start value of the Markov Chain,
128 | % nsamples = number of samples to be generated
129 | [thetaj1(i,:), acceptance_rate] = mhsample(thetaj(idx(i), :), 1, ...
130 | 'logpdf', log_posterior, ...
131 | 'proppdf', proppdf, ...
132 | 'proprnd', proprnd, ...
133 | 'thin', 3, ...
134 | 'burnin', burnin);
135 | % According to Cheung and Beck (2009) - Bayesian model updating ...,
136 | % the initial samples from reweighting and the resample of samples of
137 | % fj, in general, do not exactly follow fj1, so that the Markov
138 | % chains must "burn-in" before samples follow fj1, requiring a large
139 | % amount of samples to be generated for each level.
140 |
141 | %% Adjust the acceptance rate (optimal = 23%)
142 | % See: http://www.dms.umontreal.ca/~bedard/Beyond_234.pdf
143 | %{
144 | if acceptance_rate < 0.3
145 | % Many rejections means an inefficient chain (wasted computation
146 | %time), decrease the variance
147 | beta = 0.99*beta;
148 | elseif acceptance_rate > 0.5
149 | % High acceptance rate: Proposed jumps are very close to current
150 | % location, increase the variance
151 | beta = 1.01*beta;
152 | end
153 | %}
154 | end
155 | fprintf('\n');
156 |
157 | %% Prepare for the next iteration
158 | thetaj = thetaj1;
159 | pj = pj1;
160 | end
161 |
162 | % TMCMC provides N samples distributed according to the Posterior distribution, f(T|D)
163 | samples_fT_D = thetaj;
164 |
165 | % estimation of f(D) -- this is the normalization constant in Bayes
166 | log_fD = sum(log(S(1:j)));
167 |
168 | return; % End
169 |
170 |
171 | %% Calculate the tempering parameter p(j+1)
172 | function pj1 = calculate_pj1(log_fD_T_thetaj, pj)
173 | % find pj1 such that COV <= threshold, that is
174 | %
175 | % std(wj)
176 | % --------- <= threshold
177 | % mean(wj)
178 | %
179 | % here
180 | % size(thetaj) = N x D,
181 | % wj = fD_T(thetaj).^(pj1 - pj)
182 | % e = pj1 - pj
183 |
184 | threshold = 1; % 100% = threshold on the COV
185 |
186 | % wj = @(e) fD_T_thetaj^e; % N x 1
187 | % Note the following trick in order to calculate e:
188 | % Take into account that e>=0
189 | wj = @(e) exp(abs(e)*log_fD_T_thetaj); % N x 1
190 | %fmin = @(e) std(wj(e))/mean(wj(e)) - threshold;
191 | fmin = @(e) std(wj(e)) - threshold*mean(wj(e)) + realmin;
192 | e = abs(fzero(fmin, 0)); % e is >= 0, and fmin is an even function
193 | if isnan(e)
194 | error('There is an error finding e');
195 | end
196 |
197 | pj1 = min(1, pj + e);
198 |
199 | return; % End
200 |
201 | function proppdf = prop_pdf(x, mu, covmat, box)
202 | % This is the Proposal PDF for the Markov Chain.
203 |
204 | % Box function is the Prior PDF in the feasible region.
205 | % So if a point is out of bounds, this function will
206 | % return 0.
207 |
208 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
209 |
210 | return;
211 |
212 |
213 | function proprnd = prop_rnd(mu, covmat, box)
214 | % Sampling from the proposal PDF for the Markov Chain.
215 |
216 | while true
217 | proprnd = mvnrnd(mu, covmat, 1);
218 | if box(proprnd)
219 | % The box function is the Prior PDF in the feasible region.
220 | % If a point is out of bounds, this function will return 0 = false.
221 | break;
222 | end
223 | end
224 |
225 | return
226 |
--------------------------------------------------------------------------------
/1D Static Spring-Mass System/desktop.ini:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Adolphus8/Bayesian-Model-Updating-Tutorials/f88cbbc1bac5e4caab6f5fd620d76a25c50defa5/1D Static Spring-Mass System/desktop.ini
--------------------------------------------------------------------------------
/1D Static Spring-Mass System/log_likelihood.m:
--------------------------------------------------------------------------------
1 | function logl = log_likelihood(stiffness, modelInput, measurements, standard_deviation, ModelHandle)
2 | % Calculation of the log_likelihood for 1D Linear Static Problem:
3 | %
4 | % USAGE:
5 | % logl = log_likelihood(stiffness, displacement, measurements, standard_deviation, ModelHandle)
6 | %
7 | % INPUTS:
8 | % stiffness = epistemic parameter k [Nsamples x 1]
9 | % displacement = model input [Nobservations x 1]
10 | % measurements = experimental observations [Nobservations x 1]
11 | % standard_deviation = the standard deviation of the log likelihood function [scalar]
12 | % ModelHandle = the function handle of the model (see file "model.m")
13 | %
14 | % OUTPUTS:
15 | % logl = loglikelihood function for the set of estimated stiffness values k and
16 | % the measurements
17 |
18 |
19 | %% Evaluate the model:
20 | nchains=size(stiffness,1);
21 | %ndims=size(stiffness,2);
22 | logl=zeros(nchains,1);
23 |
24 | for n=1:nchains
25 | modelOutput = ModelHandle(stiffness(n,:),modelInput);
26 |
27 | % Note: Details to the model can be found in the file: "model.m"
28 |
29 | %% Compute the log-likelihood:
30 |
31 | logl(n) = -0.5 * (1/standard_deviation)^2 *(measurements - modelOutput)' * (measurements - modelOutput);
32 |
33 | end
34 |
35 |
--------------------------------------------------------------------------------
/1D Static Spring-Mass System/model.m:
--------------------------------------------------------------------------------
1 | function [force] = model(stiffness,displacement)
2 |
3 | %MODEL: 1-Degree of freedom mass-spring system
4 |
5 | force = - stiffness.*displacement;
6 |
7 | end
8 |
9 |
--------------------------------------------------------------------------------
/2D Eigen-value Problem/Eigenvalue_Problem.mlx:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Adolphus8/Bayesian-Model-Updating-Tutorials/f88cbbc1bac5e4caab6f5fd620d76a25c50defa5/2D Eigen-value Problem/Eigenvalue_Problem.mlx
--------------------------------------------------------------------------------
/2D Eigen-value Problem/README.md:
--------------------------------------------------------------------------------
1 | ## Instructions:
2 |
3 | * Run the tutorial MATLAB LIVE script: "Eigenvalue_Problem.mlx"
4 |
--------------------------------------------------------------------------------
/2D Eigen-value Problem/SMCsampler.m:
--------------------------------------------------------------------------------
1 | function [output] = SMCsampler(varargin)
2 | %% Sequential Monte Carlo Dynamical (SMC) sampler
3 | %
4 | % This program implements the original Sequential Monte Carlo (SMC) sampling
5 | % class (see paper by N. Chopin (2002): A sequential particle filter method
6 | % for static models - https://www.jstor.org/stable/3879283) and employs
7 | % the use of the Affine-invariant Ensemble Sampler (AIES) proposed by
8 | % Goodman and Weare (2010) to update the samples at each iteration.
9 | %
10 | % This sampler function can be employed in Sequential Bayesian Model
11 | % Updating problems involving:
12 | % - Estimating time-invariant parameter(s) via Online Bayesian Model Updating;
13 | % - Estimating time-varying parameter(s), following a recursive dynamic model;
14 | % - Predicting the time-varying parameter(s) for the next time-step given
15 | % data/observations up to the previous time-step.
16 | %
17 | %--------------------------------------------------------------------------
18 | % Author:
19 | % Adolphus Lye - adolphus.lye@liverpool.ac.uk
20 | %--------------------------------------------------------------------------
21 |
22 | % Parse the information in the name/value pairs:
23 | pnames = {'nsamples','loglikelihoods','dynamic_model',...
24 | 'priorpdf','priorrnd','burnin','lastburnin','thinchain'};
25 |
26 | % Define default values:
27 | dflts = {[], [], @(x) x, [], [], 0, 0, 3};
28 |
29 | [nsamples,loglikelihoods,dynamic_model,priorpdf,prior_rnd,...
30 | burnin,lastBurnin,thinchain] = internal.stats.parseArgs(pnames, dflts, varargin{:});
31 |
32 | %--------------------------------------------------------------------------
33 | %
34 | % Inputs:
35 | % nsamples: Scalar value of the number of samples to be generated from the Posterior;
36 | % loglikelihoods: A M x 1 cell vector of likelihood functions containing the measurements at M different time-steps;
37 | % dynamic_model: A function-handle that relates theta(t+1) and theta(t), where t is the time-step. Output is N x dim;
38 | % priorpdf: Function-handle of the Prior PDF;
39 | % prior_rnd: Function-handle of the Prior random number generator;
40 | % burnin: Number of burn-in for all iterations up to M-1;
41 | % lastBurnin: Number of burn-in for the last iteration;
42 | % stepsize: The stepsize for the Ensemble sampler in the updating step (this is the tuning parameter);
43 | % thinchain: Thin all the chains of the Ensemble sampler by only storing every k'th step (default=3);
44 | %
45 | % Outputs:
46 | % output.samples: A N x dim matrix of Posterior samples;
47 | % output.allsamples: A N x dim x (M+1) array of samples from all iterations;
48 | % output.acceptance: A M x 1 vector of acceptance rates for all iterations;
49 | % output.log_evidence: A (M+1) x 1 vector of the logarithmic of the evidence;
50 | % output.step: A M x 1 vector of step-size;
51 | % output.indicator: A M x 1 vector of indicators denoting if resampling
52 | % has occured for any iterations (1 = Yes, 0 = No);
53 | %
54 | %--------------------------------------------------------------------------
55 |
56 | %% Number of cores
57 | if ~isempty(gcp('nocreate'))
58 | pool = gcp;
59 | Ncores = pool.NumWorkers;
60 | fprintf('SMC is running on %d cores.\n', Ncores);
61 | end
62 |
63 | %% Initialize: Obtain N samples from the Prior PDF
64 |
65 | fprintf('Start SMC procedure ... \n');
66 |
67 | prior_initial = priorpdf; % Define initial Prior PDF
68 | thetaj = prior_rnd(nsamples); % theta0 = N x dim
69 | Dimensions = size(thetaj, 2); % Dimensionality of theta, dim
70 |
71 | % Initialization of matrices and vectors:
72 | thetaj1 = zeros(nsamples, Dimensions);
73 | log_evidence = zeros(size(loglikelihoods,1)+1,1); % Initiate empty vector for log evidence
74 | log_evidence(1) = 0;
75 |
76 | acceptance = zeros(size(loglikelihoods,1),1);
77 |
78 | % Samples from filter distribution, P(theta(t)|Data(1:t)):
79 | allsamples = zeros(size(thetaj,1), size(thetaj,2), size(loglikelihoods,1)+1);
80 | allsamples(:,:,1) = thetaj;
81 |
82 | % Statistics from predictive distribution, P(theta(t+1)|Data(1:t)):
83 | predictive_samples = zeros(size(thetaj,1), size(thetaj,2), size(loglikelihoods,1));
84 |
85 | % Resampling indicator vector:
86 | indicator = zeros(length(loglikelihoods), 1);
87 | % Note: This indicator vector returns a 1 for the iteration(s) where
88 | % resampling is initiated and 0 otherwise.
89 |
90 | %% Main sampling loop
91 | for iter = 1:length(loglikelihoods)
92 |
93 | fprintf('SMC: Iteration j = %2d \n', iter);
94 |
95 | loglikelihood = loglikelihoods{iter};
96 |
97 | % Compute loglikelihood values for each sample:
98 | logL = zeros(nsamples,1);
99 | for l = 1:nsamples
100 | logL(l) = loglikelihood(thetaj(l,:));
101 | end
102 |
103 | % Error check:
104 | if any(isinf(logL))
105 | error('The prior distribution is too far from the true region');
106 | end
107 |
108 | %% Compute weights of the samples, wj:
109 |
110 | % To compute the nominal weights:
111 | fprintf('Computing the weights ...\n');
112 | wj = exp(logL);
113 |
114 | % To compute the log evidence for the current iteration:
115 | log_evidence(iter+1) = log(mean(wj)) + log_evidence(iter);
116 |
117 | % Check step for wj:
118 | for i = 1:nsamples
119 | if wj(i) == 0
120 | wj(i) = 1e-100;
121 | end
122 | end
123 |
124 | wj_norm = wj./sum(wj); % To normalise the weights
125 |
126 | %% Check step - Compute the sum of wj_norm and see if it is < nsamples/2:
127 |
128 | fprintf('Computing effective sample size ... \n');
129 | Neff = 1/(sum(wj_norm.^2));
130 | threshold = nsamples/2;
131 |
132 | %% Resampling step (conditional if Neff < threshold):
133 |
134 | if Neff < threshold
135 | fprintf('Resampling step initiated ... \n');
136 |
137 | dx = randsample(nsamples, nsamples, true, wj_norm);
138 |
139 | thetaj_resampled = zeros(nsamples, Dimensions);
140 | for d = 1:nsamples
141 | thetaj_resampled(d,:) = thetaj(dx(d),:);
142 | end
143 |
144 | thetaj = thetaj_resampled;
145 | wj_norm = (1/nsamples).*ones(nsamples,1);
146 | indicator(iter) = 1;
147 |
148 | end
149 |
150 | %% Update the samples according to the current Posterior using MH sampler:
151 |
152 | % Define the logposterior:
153 | log_posterior = @(x) log(priorpdf(x)) + loglikelihood(x);
154 |
155 | % Weighted mean for Proposal distribution
156 | mu = zeros(1, Dimensions);
157 | for l = 1:nsamples
158 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
159 | end
160 |
161 | % Covariance matrix for Proposal distribution:
162 | cov_gauss = zeros(Dimensions);
163 | for k = 1:nsamples
164 | tk_mu = thetaj(k,:) - mu;
165 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
166 | end
167 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
168 |
169 | % Define the Proposal distribution:
170 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, prior_initial); % q(x,y) = q(x|y).
171 | proprnd = @(x) prop_rnd(x, cov_gauss, prior_initial);
172 |
173 | if iter == length(loglikelihoods)
174 | burnin = lastBurnin;
175 | end
176 |
177 | %% Start N different Markov chains
178 | fprintf('Markov chains ...\n\n');
179 |
180 | idx = randsample(nsamples, nsamples, true, wj_norm);
181 | for i = 1:nsamples % For parallel, type: parfor
182 | [thetaj1(i,:), acceptance_rate(i)] = mhsample(thetaj(idx(i),:), 1, ...
183 | 'logpdf', log_posterior, ...
184 | 'proppdf', proppdf, ...
185 | 'proprnd', proprnd, ...
186 | 'thin', thinchain, ...
187 | 'burnin', burnin);
188 | end
189 | fprintf('\n');
190 | acceptance(iter) = mean(acceptance_rate); % To store the acceptance rate values
191 |
192 | %% Prediction step:
193 |
194 | % Define the Predictive distribution of the samples, P(theta(t+1)|Data(t)):
195 | predictive_samples(:,:,iter) = dynamic_model(thetaj1);
196 |
197 | % Compute the Bandwidth vector for the kernel density function:
198 | pred_samps = predictive_samples(:,:,iter);
199 |
200 | bw = zeros(Dimensions,1);
201 | for dim = 1:Dimensions
202 | bw(dim) = std(pred_samps(:,dim)) .* (4/((Dimensions + 2) .* nsamples)).^(1/(Dimensions + 4));
203 | end
204 |
205 | % Define the Predictive PDF, P(theta(t+1)|Data(t)) using mvksdensity:
206 | pred_pdf = @(x) mvksdensity(pred_samps, x, 'Bandwidth', bw);
207 |
208 | %% Prepare for the next iteration:
209 |
210 | allsamples(:,:,iter+1) = thetaj1;
211 | thetaj = pred_samps;
212 | priorpdf = @(x) pred_pdf(x);
213 |
214 | end
215 |
216 | %% Description of outputs:
217 |
218 | output.samples = thetaj; % To only show samples from the final filter distribution
219 | output.allsamples = allsamples; % To only show all filter samples across all iterations
220 | output.prediction = predictive_samples; % To only show all prediction samples across all iterations
221 | output.acceptance = acceptance; % To show the mean acceptance rates for all iterations
222 | output.log_evidence = log_evidence; % To show the (M+1) x 1 vector of the logarithmic of the evidence;
223 | output.indicator = indicator; % To indicate the iterations whereby resampling took place (denoted by 1s).
224 |
225 | fprintf('End of SMC procedure. \n\n');
226 |
227 | return; % End
228 |
229 | function proppdf = prop_pdf(x, mu, covmat, box)
230 | % This is the Proposal PDF for the Markov Chain.
231 |
232 | % Box function is the Prior PDF in the feasible region.
233 | % So if a point is out of bounds, this function will
234 | % return 0.
235 |
236 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
237 |
238 | return;
239 |
240 |
241 | function proprnd = prop_rnd(mu, covmat, box)
242 | % Sampling from the proposal PDF for the Markov Chain.
243 |
244 | while true
245 | proprnd = mvnrnd(mu, covmat, 1);
246 | if box(proprnd)
247 | % The box function is the Prior PDF in the feasible region.
248 | % If a point is out of bounds, this function will return 0 = false.
249 | break;
250 | end
251 | end
252 |
253 | return
--------------------------------------------------------------------------------
/2D Eigen-value Problem/TMCMCsampler.m:
--------------------------------------------------------------------------------
1 | function [samples_fT_D, log_fD] = TMCMCsampler(varargin)
2 | %% Transitional Markov Chain Monte Carlo sampler
3 | %
4 | % This program implements a method described in:
5 | % Ching, J. and Chen, Y. (2007). "Transitional Markov Chain Monte Carlo
6 | % Method for Bayesian Model Updating, Model Class Selection, and Model
7 | % Averaging." J. Eng. Mech., 133(7), 816-832.
8 | %
9 | % Usage:
10 | % [samples_fT_D, fD] = tmcmc_v1(fD_T, fT, sample_from_fT, N);
11 | %
12 | % where:
13 | %
14 | % inputs:
15 | % log_fD_T = function handle of log(fD_T(t)), Loglikelihood
16 | %
17 | % fT = function handle of fT(t), Prior PDF
18 | %
19 | % sample_from_fT = handle to a function that samples from of fT(t),
20 | % Sampling rule function from Prior PDF
21 | %
22 | % nsamples = number of samples of fT_D, Posterior, to generate
23 | %
24 | % outputs:
25 | % samples_fT_D = samples of fT_D (N x D) = samples from Posterior
26 | % distribution
27 | %
28 | % log_fD = log(evidence) = log(normalization constant)
29 |
30 | % ------------------------------------------------------------------------
31 | % who when observations
32 | %--------------------------------------------------------------------------
33 | % Diego Andres Alvarez Jul-24-2013 First algorithm
34 | %--------------------------------------------------------------------------
35 | % Diego Andres Alvarez - daalvarez@unal.edu.co
36 | % Edoardo Patelli - edoardo.patelli@strath.ac.uk
37 |
38 | % parse the information in the name/value pairs:
39 | pnames = {'nsamples','loglikelihood','priorpdf','priorrnd','burnin','lastburnin','beta'};
40 |
41 | dflts = {[],[],[],[],[],0,0.2}; % define default values
42 |
43 | [nsamples,loglikelihood,priorpdf,prior_rnd,burnin,lastBurnin,beta] = ...
44 | internal.stats.parseArgs(pnames, dflts, varargin{:});
45 |
46 | %% Obtain N samples from the prior pdf f(T)
47 | j = 0; % Initialise loop for the transitional likelihood
48 | thetaj = prior_rnd(nsamples); % theta0 = N x D
49 | pj = 0; % p0 = 0 (initial tempering parameter)
50 | Dimensions = size(thetaj, 2); % size of the vector theta
51 |
52 | %% Initialization of matrices and vectors
53 | thetaj1 = zeros(nsamples, Dimensions);
54 | %log_fD_T_thetaj = zeros(nsamples,1);
55 |
56 | %% Main loop
57 | while pj < 1
58 | j = j+1;
59 |
60 | %% Calculate the tempering parameter p(j+1):
61 | for l = 1:nsamples
62 | log_fD_T_thetaj(l) = loglikelihood(thetaj(l,:));
63 | end
64 | if any(isinf(log_fD_T_thetaj))
65 | error('The prior distribution is too far from the true region');
66 | end
67 | pj1 = calculate_pj1(log_fD_T_thetaj, pj);
68 | fprintf('TMCMC: Iteration j = %2d, pj1 = %f\n', j, pj1);
69 |
70 | %% Compute the plausibility weight for each sample wrt f_{j+1}
71 | fprintf('Computing the weights ...\n');
72 | % wj = fD_T(thetaj).^(pj1-pj); % N x 1 (eq 12)
73 | a = (pj1-pj)*log_fD_T_thetaj;
74 | wj = exp(a);
75 | wj_norm = wj./sum(wj); % normalization of the weights
76 |
77 | %% Compute S(j) = E[w{j}] (eq 15)
78 | S(j) = mean(wj);
79 |
80 | %% Do the resampling step to obtain N samples from f_{j+1}(theta) and
81 | % then perform Metropolis-Hastings on each of these samples using as a
82 | % stationary PDF "fj1"
83 | % fj1 = @(t) fT(t).*log_fD_T(t).^pj1; % stationary PDF (eq 11) f_{j+1}(theta)
84 | log_posterior = @(t) log(priorpdf(t)) + pj1*loglikelihood(t);
85 |
86 |
87 | % and using as proposal PDF a Gaussian centered at thetaj(idx,:) and
88 | % with covariance matrix equal to an scaled version of the covariance
89 | % matrix of fj1:
90 |
91 | % weighted mean
92 | mu = zeros(1, Dimensions);
93 | for l = 1:nsamples
94 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
95 | end
96 |
97 | % scaled covariance matrix of fj1 (eq 17)
98 | cov_gauss = zeros(Dimensions);
99 | for k = 1:nsamples
100 | % this formula is slightly different to eq 17 (the transpose)
101 | % because of the size of the vectors)m and because Ching and Chen
102 | % forgot to normalize the weight wj:
103 | tk_mu = thetaj(k,:) - mu;
104 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
105 | end
106 | cov_gauss = beta^2 * cov_gauss;
107 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
108 |
109 | % Define the Proposal distribution:
110 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, priorpdf); %q(x,y) = q(x|y).
111 | proprnd = @(x) prop_rnd(x, cov_gauss, priorpdf);
112 |
113 | %% During the last iteration we require to do a better burnin in order
114 | % to guarantee the quality of the samples:
115 | if pj1 == 1
116 | burnin = lastBurnin;
117 | end;
118 |
119 | %% Start N different Markov chains
120 | fprintf('Markov chains ...\n\n');
121 | idx = randsample(nsamples, nsamples, true, wj_norm);
122 | for i = 1:nsamples % For parallel, type: parfor
123 | %% Sample one point with probability wj_norm
124 |
125 | % smpl = mhsample(start, nsamples,
126 | % 'pdf', pdf, 'proppdf', proppdf, 'proprnd', proprnd);
127 | % start = row vector containing the start value of the Markov Chain,
128 | % nsamples = number of samples to be generated
129 | [thetaj1(i,:), acceptance_rate] = mhsample(thetaj(idx(i), :), 1, ...
130 | 'logpdf', log_posterior, ...
131 | 'proppdf', proppdf, ...
132 | 'proprnd', proprnd, ...
133 | 'thin', 3, ...
134 | 'burnin', burnin);
135 | % According to Cheung and Beck (2009) - Bayesian model updating ...,
136 | % the initial samples from reweighting and the resample of samples of
137 | % fj, in general, do not exactly follow fj1, so that the Markov
138 | % chains must "burn-in" before samples follow fj1, requiring a large
139 | % amount of samples to be generated for each level.
140 |
141 | %% Adjust the acceptance rate (optimal = 23%)
142 | % See: http://www.dms.umontreal.ca/~bedard/Beyond_234.pdf
143 | %{
144 | if acceptance_rate < 0.3
145 | % Many rejections means an inefficient chain (wasted computation
146 | %time), decrease the variance
147 | beta = 0.99*beta;
148 | elseif acceptance_rate > 0.5
149 | % High acceptance rate: Proposed jumps are very close to current
150 | % location, increase the variance
151 | beta = 1.01*beta;
152 | end
153 | %}
154 | end
155 | fprintf('\n');
156 |
157 | %% Prepare for the next iteration
158 | thetaj = thetaj1;
159 | pj = pj1;
160 | end
161 |
162 | % TMCMC provides N samples distributed according to the Posterior distribution, f(T|D)
163 | samples_fT_D = thetaj;
164 |
165 | % estimation of f(D) -- this is the normalization constant in Bayes
166 | log_fD = sum(log(S(1:j)));
167 |
168 | return; % End
169 |
170 |
171 | %% Calculate the tempering parameter p(j+1)
172 | function pj1 = calculate_pj1(log_fD_T_thetaj, pj)
173 | % find pj1 such that COV <= threshold, that is
174 | %
175 | % std(wj)
176 | % --------- <= threshold
177 | % mean(wj)
178 | %
179 | % here
180 | % size(thetaj) = N x D,
181 | % wj = fD_T(thetaj).^(pj1 - pj)
182 | % e = pj1 - pj
183 |
184 | threshold = 1; % 100% = threshold on the COV
185 |
186 | % wj = @(e) fD_T_thetaj^e; % N x 1
187 | % Note the following trick in order to calculate e:
188 | % Take into account that e>=0
189 | wj = @(e) exp(abs(e)*log_fD_T_thetaj); % N x 1
190 | %fmin = @(e) std(wj(e))/mean(wj(e)) - threshold;
191 | fmin = @(e) std(wj(e)) - threshold*mean(wj(e)) + realmin;
192 | e = abs(fzero(fmin, 0)); % e is >= 0, and fmin is an even function
193 | if isnan(e)
194 | error('There is an error finding e');
195 | end
196 |
197 | pj1 = min(1, pj + e);
198 |
199 | return; % End
200 |
201 | function proppdf = prop_pdf(x, mu, covmat, box)
202 | % This is the Proposal PDF for the Markov Chain.
203 |
204 | % Box function is the Prior PDF in the feasible region.
205 | % So if a point is out of bounds, this function will
206 | % return 0.
207 |
208 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
209 |
210 | return;
211 |
212 |
213 | function proprnd = prop_rnd(mu, covmat, box)
214 | % Sampling from the proposal PDF for the Markov Chain.
215 |
216 | while true
217 | proprnd = mvnrnd(mu, covmat, 1);
218 | if box(proprnd)
219 | % The box function is the Prior PDF in the feasible region.
220 | % If a point is out of bounds, this function will return 0 = false.
221 | break;
222 | end
223 | end
224 |
225 | return
226 |
--------------------------------------------------------------------------------
/2D Eigen-value Problem/desktop.ini:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Adolphus8/Bayesian-Model-Updating-Tutorials/f88cbbc1bac5e4caab6f5fd620d76a25c50defa5/2D Eigen-value Problem/desktop.ini
--------------------------------------------------------------------------------
/2D Eigen-value Problem/log_likelihood.m:
--------------------------------------------------------------------------------
1 | function logl = log_likelihood(Thetas, Eigenvalues, standard_deviations, ModelHandle)
2 | % Calculation of the log_likelihood for 2D Inverse Eigenvalue Problem:
3 | %
4 | % USAGE:
5 | % logl = log_likelihood(Thetas, Eigenvalues, standard_deviations, ModelHandle)
6 | %
7 | % INPUTS:
8 | % Thetas = Vector of epistemic parameters theta_1 and theta_2 [Nsamples x 2]
9 | % Eigenvalues = Observations of Eigenvalues from the 2 x 2 Square Matrix [Nobservations x 2]
10 | % standard_deviations = Vector of the standard deviations of the 2D log likelihood function [2 x 1]
11 | % ModelHandle = the function handle of the model (see file "model.m")
12 | %
13 | % OUTPUTS:
14 | % logl = loglikelihood function for the set of estimated values of theta_1
15 | % and theta_2 as well as the Eigenvalues
16 |
17 |
18 | %% Evaluate the model:
19 | nchains=size(Thetas,1);
20 | logl=zeros(nchains,1);
21 |
22 | for n=1:nchains
23 | modelOutput = ModelHandle(Thetas(n,:));
24 |
25 | % Note: Details to the model can be found in the file: "model.m"
26 |
27 | %% Compute the overall 2D log-likelihood:
28 |
29 | logl_1 = - 0.5 * (1/standard_deviations(1))^2 *(Eigenvalues(:,1) - modelOutput(1))' * (Eigenvalues(:,1) - modelOutput(1));
30 |
31 | % Note: logl_1 is the loglikelihood function in the theta_1 dimension.
32 |
33 | logl_2 = - 0.5 * (1/standard_deviations(2))^2 *(Eigenvalues(:,2) - modelOutput(2))' * (Eigenvalues(:,2) - modelOutput(2));
34 |
35 | % Note: logl_2 is the loglikelihood function in the theta_2 dimension.
36 |
37 | logl(n) = logl_1 + logl_2;
38 |
39 | % logl is the overall 2D loglikelihood function.
40 |
41 | end
--------------------------------------------------------------------------------
/2D Eigen-value Problem/model.m:
--------------------------------------------------------------------------------
1 | function [eigenvalues] = model(Thetas)
2 |
3 | %MODEL: Eigenvalues of 2 x 2 square matrix
4 |
5 | eigenvalues(:,1) = 0.5*((Thetas(:,1) + 2*Thetas(:,2)) + (Thetas(:,1).^2 + 4*(Thetas(:,2).^2)).^0.5);
6 |
7 | eigenvalues(:,2) = 0.5*((Thetas(:,1) + 2*Thetas(:,2)) - (Thetas(:,1).^2 + 4*(Thetas(:,2).^2)).^0.5);
8 |
9 | end
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/DenHartogHarmonic.m:
--------------------------------------------------------------------------------
1 | function [output] = DenHartogHarmonic(beta_v)
2 | %% Function-handle for Analytical plots of Tramissibility and Phase Angles:
3 | %
4 | % This set-up is based on the Single Degree-of-Freedom Dynamical System
5 | % subjected to Coulomb Friction Force that is presented in the literature:
6 | %
7 | % L. Marino and A. Cicirello (2020). Experimental investigation of a single-
8 | % degree-of-freedom system with Coulomb friction. Nonlinear Dynamics, 99(3),
9 | % 1781-1799. doi: 10.1007/s11071-019-05443-2
10 | %
11 | % This set-up is applicable only for the Base motion (with fixed wall) case.
12 | %-------------------------------------------------------------------------%
13 | %% Define key parameters:
14 |
15 | % Define the non-dimensional parameters:
16 | r_v = [0.01:0.01:0.9 0.901:0.001:0.999 1.001:0.001:1.099 1.1:0.025:2.6];
17 | beta_v = [0 beta_v];
18 |
19 | dt = pi/15; % Time-step
20 | t_half = 0:dt:pi; % Time half-period
21 | N_cyc = 30; % To be selected for FFT performance
22 |
23 | %% Obtain the boundary values of Displacement Transmissibility and Phase Angles:
24 |
25 | for ir = 1:length(r_v) % For each Frequency ratio value
26 | r = r_v(ir);
27 |
28 | % Response and damping functions:
29 | U = sin(pi/r)/(r*(1+cos(pi/r))); % See Eq. (7)
30 | V = 1/(1-r^2); % See Eq. (8)
31 |
32 | % Boundaries:
33 | sr = (r*sin(t_half/r)+U*r^2*(cos(t_half)-cos(t_half/r)))./sin(t_half); % See Eq. (9)
34 | S = max(sr(1:end-1));
35 |
36 | beta_bound = sqrt(V^2/(U^2+(S/r^2)^2)); % See Eq. (6)
37 | X_bound = sqrt(V^2 - beta_bound^2*U^2); % Displacement Tranmissibility bound (see Eq. (11))
38 | ph_bound = angleCalc(- beta_bound*U/V, X_bound/V, 'rad'); % Compute the bounds for Phase Angles
39 |
40 | x_bound_half = X_bound*cos(t_half) + beta_bound*U*sin(t_half) + ...
41 | beta_bound*(1 - cos(t_half/r) - U*r*sin(t_half/r)); % See Eq. (15)
42 | y_bound_half = cos(t_half + ph_bound); % See Eq. (16)
43 |
44 | t_period = [t_half t_half(2:end)+pi];
45 | x_bound_period = [x_bound_half -x_bound_half(2:end)];
46 | y_bound_period = [y_bound_half -y_bound_half(2:end)];
47 |
48 | t = t_period;
49 | for in = 1:N_cyc - 1
50 | t = [t t_period(2:end)+2*in*pi];
51 | end
52 |
53 | x_bound = [x_bound_period repmat(x_bound_period(2:end),1,N_cyc-1)];
54 | y_bound = [y_bound_period repmat(y_bound_period(2:end),1,N_cyc-1)];
55 |
56 | %% Fast-Fourier Transform Step:
57 |
58 | Nt = length(t); % No. of time-steps
59 | fs = 1/dt; % Frequency of data-collection
60 | df = 1/t(end); % Frequency steps
61 | omega = 2*pi*(0:df:fs); % Value of omegas = 2*pi*f
62 |
63 | % Compute the bounds in the time and frequency domains:
64 | x_bound_fft = fft(x_bound)/Nt;
65 | y_bound_fft = fft(y_bound)/Nt;
66 |
67 | i_peak = find(round(omega,3)>=1,1);
68 | % Note: round(omega,3) rounds the omega value to the nearest thousandth (10^-3).
69 | X_bound_peak = abs(x_bound_fft(i_peak));
70 | Y_bound_peak = abs(y_bound_fft(i_peak));
71 |
72 | % Identify the Den-Hartog's bound for Transmissibility (see Eq. (29)):
73 | Tr_bound(ir) = X_bound_peak/Y_bound_peak;
74 |
75 | % Identify the Den-Hartog's bound for Phase Angles (see Eq. (30)):
76 | phase_bound(ir) = rad2deg(angle(y_bound_fft(i_peak)/x_bound_fft(i_peak)));
77 |
78 | % Compute the Transmissibility and Phase Angle Response for different beta values:
79 | for ib = 1:length(beta_v)
80 | beta = beta_v(ib);
81 |
82 | if beta <= beta_bound || beta == 0 % In the continuous motion condition
83 | X = sqrt(V^2 - (beta^2 * U^2)); % See Eq. (11)
84 | ph = angleCalc(-beta*U/V,X/V,'rad'); % See Eq. (12)
85 | x_half = X*cos(t_half)+beta*U*sin(t_half)+beta*(1-cos(t_half/r)-U*r*sin(t_half/r)); % See Eq. (15)
86 | y_half = cos(t_half+ph); % See Eq. (16)
87 |
88 | % Post-processing step:
89 | x_period = [x_half -x_half(2:end)];
90 | y_period = [y_half -y_half(2:end)];
91 |
92 | x = [x_period repmat(x_period(2:end),1,N_cyc-1)];
93 | y = [y_period repmat(y_period(2:end),1,N_cyc-1)];
94 |
95 | y_fft = fft(y)/Nt;
96 | x_fft = fft(x)/Nt;
97 |
98 | Y_peak = abs(y_fft(i_peak));
99 | X_peak = abs(x_fft(i_peak));
100 |
101 | % Transimissibility response for r = r_v and beta = beta_v (see Eq. (29)):
102 | Tr(ib,ir) = X_peak/Y_peak;
103 |
104 | % Phase Angle response for r = r_v and beta = beta_v (see Eq. (30)):
105 | phase(ib,ir) = rad2deg(angle(y_fft(i_peak)/x_fft(i_peak)));
106 |
107 | else % In the stick-slip domain
108 | Tr(ib,ir) = NaN;
109 | phase(ib,ir) = NaN;
110 | end
111 |
112 | end
113 | end
114 |
115 | %% Consolidate the output:
116 | output.frequency_ratios = r_v;
117 | output.trans = Tr;
118 | output.phase_angles = phase;
119 | output.trans_bound = Tr_bound;
120 | output.phase_bound = phase_bound;
121 | end
122 |
123 | function theta = angleCalc(S,C,out_mode)
124 | %% This function computes the angle from sin and cos values (-180,180] degree.
125 | %
126 | % Usage:
127 | % theta = angleCalc(S,C,out_mode)
128 | %
129 | % Input:
130 | % S: Sine value of the angle
131 | % C: Cosine value of the angle
132 | % out_mode: 'deg' OR 'rad'
133 | % Note: default output mode is in degree
134 | %
135 | % Output:
136 | % theta: Angles in degrees or radians.
137 | %
138 | % Example:
139 | % theta = angleCalc(sin(-2*pi/3),cos(-2*pi/3))
140 | % theta = -120;
141 | % theta= angleCalc(sin(2*pi/3),cos(2*pi/3),'rad')
142 | % theta= 2.0944 [rad]
143 | % --------------Disi A Jun 25, 2013
144 | %% Define the function:
145 | if nargin < 3
146 | out_mode='deg';
147 | end
148 |
149 | if strcmp(out_mode,'deg')
150 | cons = 180/pi;
151 | else
152 | cons = 1;
153 | end
154 |
155 | for i = 1:length(S)
156 | theta(i) = asin(S(i));
157 | if C(i) < 0
158 | if S(i) > 0
159 | theta(i) = pi - theta(i);
160 | elseif S(i) < 0
161 | theta(i) = - pi - theta(i);
162 | else % If S(i) = 0
163 | theta(i) = theta(i) + pi;
164 | end
165 | end
166 |
167 | theta(i) = theta(i) .* cons;
168 | end
169 | end
170 |
171 |
172 |
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/TMCMCsampler.m:
--------------------------------------------------------------------------------
1 | function [output] = TMCMCsampler(varargin)
2 | %% Transitional Markov Chain Monte Carlo sampler
3 | %
4 | % This program implements a method described in:
5 | % Ching, J. and Chen, Y. (2007). "Transitional Markov Chain Monte Carlo
6 | % Method for Bayesian Model Updating, Model Class Selection, and Model
7 | % Averaging." J. Eng. Mech., 133(7), 816-832.
8 | %
9 | % Usage:
10 | % [samples_fT_D, fD] = tmcmc_v1(fD_T, fT, sample_from_fT, N);
11 | %
12 | % where:
13 | %
14 | % inputs:
15 | % log_fD_T = function handle of log(fD_T(t)), Loglikelihood
16 | %
17 | % fT = function handle of fT(t), Prior PDF
18 | %
19 | % sample_from_fT = handle to a function that samples from of fT(t),
20 | % Sampling rule function from Prior PDF
21 | %
22 | % nsamples = number of samples of fT_D, Posterior, to generate
23 | %
24 | % outputs:
25 | % samples_fT_D = samples of fT_D (N x D) = samples from Posterior
26 | % distribution
27 | %
28 | % log_fD = log(evidence) = log(normalization constant)
29 |
30 | % ------------------------------------------------------------------------
31 | % who when observations
32 | %--------------------------------------------------------------------------
33 | % Diego Andres Alvarez Jul-24-2013 First algorithm
34 | %--------------------------------------------------------------------------
35 | % Diego Andres Alvarez - daalvarez@unal.edu.co
36 | % Edoardo Patelli - edoardo.patelli@strath.ac.uk
37 |
38 | % parse the information in the name/value pairs:
39 | pnames = {'nsamples','loglikelihood','priorpdf','priorrnd','burnin','lastburnin','beta'};
40 |
41 | dflts = {[],[],[],[],[],0,0.2}; % define default values
42 |
43 | [nsamples,loglikelihood,priorpdf,prior_rnd,burnin,lastBurnin,beta] = ...
44 | internal.stats.parseArgs(pnames, dflts, varargin{:});
45 |
46 | %% Obtain N samples from the prior pdf f(T)
47 | j = 0; % Initialise loop for the transitional likelihood
48 | thetaj = prior_rnd(nsamples); % theta0 = N x D
49 | pj = 0; % p0 = 0 (initial tempering parameter)
50 | Dimensions = size(thetaj, 2); % size of the vector theta
51 |
52 | count = 1; % Counter
53 | samps(:,:,count) = thetaj;
54 | beta_j(count) = pj;
55 |
56 | %% Initialization of matrices and vectors
57 | thetaj1 = zeros(nsamples, Dimensions);
58 | %log_fD_T_thetaj = zeros(nsamples,1);
59 |
60 | %% Main loop
61 | while pj < 1
62 | j = j+1;
63 |
64 | %% Calculate the tempering parameter p(j+1):
65 | for l = 1:nsamples
66 | log_fD_T_thetaj(l) = loglikelihood(thetaj(l,:));
67 | end
68 | if any(isinf(log_fD_T_thetaj))
69 | error('The prior distribution is too far from the true region');
70 | end
71 | pj1 = calculate_pj1(log_fD_T_thetaj, pj);
72 | fprintf('TMCMC: Iteration j = %2d, pj1 = %f\n', j, pj1);
73 |
74 | %% Compute the plausibility weight for each sample wrt f_{j+1}
75 | fprintf('Computing the weights ...\n');
76 | % wj = fD_T(thetaj).^(pj1-pj); % N x 1 (eq 12)
77 | a = (pj1-pj)*log_fD_T_thetaj;
78 | wj = exp(a);
79 | wj_norm = wj./sum(wj); % normalization of the weights
80 |
81 | %% Compute S(j) = E[w{j}] (eq 15)
82 | S(j) = mean(wj);
83 |
84 | %% Do the resampling step to obtain N samples from f_{j+1}(theta) and
85 | % then perform Metropolis-Hastings on each of these samples using as a
86 | % stationary PDF "fj1"
87 | % fj1 = @(t) fT(t).*log_fD_T(t).^pj1; % stationary PDF (eq 11) f_{j+1}(theta)
88 | log_posterior = @(t) log(priorpdf(t)) + pj1*loglikelihood(t);
89 |
90 |
91 | % and using as proposal PDF a Gaussian centered at thetaj(idx,:) and
92 | % with covariance matrix equal to an scaled version of the covariance
93 | % matrix of fj1:
94 |
95 | % weighted mean
96 | mu = zeros(1, Dimensions);
97 | for l = 1:nsamples
98 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
99 | end
100 |
101 | % scaled covariance matrix of fj1 (eq 17)
102 | cov_gauss = zeros(Dimensions);
103 | for k = 1:nsamples
104 | % this formula is slightly different to eq 17 (the transpose)
105 | % because of the size of the vectors)m and because Ching and Chen
106 | % forgot to normalize the weight wj:
107 | tk_mu = thetaj(k,:) - mu;
108 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
109 | end
110 | cov_gauss = beta^2 * cov_gauss;
111 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
112 |
113 | % Define the Proposal distribution:
114 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, priorpdf); %q(x,y) = q(x|y).
115 | proprnd = @(x) prop_rnd(x, cov_gauss, priorpdf);
116 |
117 | %% During the last iteration we require to do a better burnin in order
118 | % to guarantee the quality of the samples:
119 | if pj1 == 1
120 | burnin = lastBurnin;
121 | end;
122 |
123 | %% Start N different Markov chains
124 | fprintf('Markov chains ...\n\n');
125 | idx = randsample(nsamples, nsamples, true, wj_norm);
126 |
127 | for i = 1:nsamples % For parallel, type: parfor
128 | %% Sample one point with probability wj_norm
129 |
130 | % smpl = mhsample(start, nsamples,
131 | % 'pdf', pdf, 'proppdf', proppdf, 'proprnd', proprnd);
132 | % start = row vector containing the start value of the Markov Chain,
133 | % nsamples = number of samples to be generated
134 | [thetaj1(i,:), acceptance_rate(i)] = mhsample(thetaj(idx(i), :), 1, ...
135 | 'logpdf', log_posterior, ...
136 | 'proppdf', proppdf, ...
137 | 'proprnd', proprnd, ...
138 | 'thin', 3, ...
139 | 'burnin', burnin);
140 | % According to Cheung and Beck (2009) - Bayesian model updating ...,
141 | % the initial samples from reweighting and the resample of samples of
142 | % fj, in general, do not exactly follow fj1, so that the Markov
143 | % chains must "burn-in" before samples follow fj1, requiring a large
144 | % amount of samples to be generated for each level.
145 |
146 | %% Adjust the acceptance rate (optimal = 23%)
147 | % See: http://www.dms.umontreal.ca/~bedard/Beyond_234.pdf
148 | %{
149 | if acceptance_rate < 0.3
150 | % Many rejections means an inefficient chain (wasted computation
151 | %time), decrease the variance
152 | beta = 0.99*beta;
153 | elseif acceptance_rate > 0.5
154 | % High acceptance rate: Proposed jumps are very close to current
155 | % location, increase the variance
156 | beta = 1.01*beta;
157 | end
158 | %}
159 | end
160 | fprintf('\n');
161 | acceptance(count) = mean(acceptance_rate);
162 | %% Prepare for the next iteration
163 | count = count+1;
164 | samps(:,:,count) = thetaj1;
165 | thetaj = thetaj1;
166 | pj = pj1;
167 | beta_j(count) = pj;
168 | end
169 |
170 | % estimation of f(D) -- this is the normalization constant in Bayes
171 | log_fD = sum(log(S(1:j)));
172 |
173 | %% Description of outputs:
174 |
175 | output.allsamples = samps; % To show samples from all transitional distributions
176 | output.samples = samps(:,:,end); % To only show samples from the final posterior
177 | output.log_evidence = log_fD; % To generate the logarithmic of the evidence
178 | output.acceptance = acceptance; % To show the mean acceptance rates for all iterations
179 | output.beta = beta_j; % To show the values of temepring parameters, beta_j
180 |
181 | return; % End
182 |
183 |
184 | %% Calculate the tempering parameter p(j+1)
185 | function pj1 = calculate_pj1(log_fD_T_thetaj, pj)
186 | % find pj1 such that COV <= threshold, that is
187 | %
188 | % std(wj)
189 | % --------- <= threshold
190 | % mean(wj)
191 | %
192 | % here
193 | % size(thetaj) = N x D,
194 | % wj = fD_T(thetaj).^(pj1 - pj)
195 | % e = pj1 - pj
196 |
197 | threshold = 1; % 100% = threshold on the COV
198 |
199 | % wj = @(e) fD_T_thetaj^e; % N x 1
200 | % Note the following trick in order to calculate e:
201 | % Take into account that e>=0
202 | wj = @(e) exp(abs(e)*log_fD_T_thetaj); % N x 1
203 | %fmin = @(e) std(wj(e))/mean(wj(e)) - threshold;
204 | fmin = @(e) std(wj(e)) - threshold*mean(wj(e)) + realmin;
205 | e = abs(fzero(fmin, 0)); % e is >= 0, and fmin is an even function
206 | if isnan(e)
207 | error('There is an error finding e');
208 | end
209 |
210 | pj1 = min(1, pj + e);
211 |
212 | return; % End
213 |
214 | function proppdf = prop_pdf(x, mu, covmat, box)
215 | % This is the Proposal PDF for the Markov Chain.
216 |
217 | % Box function is the Prior PDF in the feasible region.
218 | % So if a point is out of bounds, this function will
219 | % return 0.
220 |
221 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
222 |
223 | return;
224 |
225 |
226 | function proprnd = prop_rnd(mu, covmat, box)
227 | % Sampling from the proposal PDF for the Markov Chain.
228 |
229 | while true
230 | proprnd = mvnrnd(mu, covmat, 1);
231 | if box(proprnd)
232 | % The box function is the Prior PDF in the feasible region.
233 | % If a point is out of bounds, this function will return 0 = false.
234 | break;
235 | end
236 | end
237 |
238 | return
239 |
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/TMCMCsampler2.m:
--------------------------------------------------------------------------------
1 | function [output] = TMCMCsampler2(varargin)
2 | %% Transitional Markov Chain Monte Carlo sampler
3 | %
4 | % This program implements a method described in:
5 | % Ching, J. and Chen, Y. (2007). "Transitional Markov Chain Monte Carlo
6 | % Method for Bayesian Model Updating, Model Class Selection, and Model
7 | % Averaging." J. Eng. Mech., 133(7), 816-832.
8 | %
9 | % Usage:
10 | % [samples_fT_D, fD] = tmcmc_v1(fD_T, fT, sample_from_fT, N);
11 | %
12 | % where:
13 | %
14 | % inputs:
15 | % log_fD_T = function handle of log(fD_T(t)), Loglikelihood
16 | %
17 | % fT = function handle of fT(t), Prior PDF
18 | %
19 | % sample_from_fT = handle to a function that samples from of fT(t),
20 | % Sampling rule function from Prior PDF
21 | %
22 | % nsamples = number of samples of fT_D, Posterior, to generate
23 | %
24 | % outputs:
25 | % samples_fT_D = samples of fT_D (N x D) = samples from Posterior
26 | % distribution
27 | %
28 | % log_fD = log(evidence) = log(normalization constant)
29 |
30 | % ------------------------------------------------------------------------
31 | % who when observations
32 | %--------------------------------------------------------------------------
33 | % Diego Andres Alvarez Jul-24-2013 First algorithm
34 | %--------------------------------------------------------------------------
35 | % Diego Andres Alvarez - daalvarez@unal.edu.co
36 | % Edoardo Patelli - edoardo.patelli@strath.ac.uk
37 |
38 | % parse the information in the name/value pairs:
39 | pnames = {'nsamples','loglikelihood','priorpdf','priorrnd','burnin','lastburnin','beta'};
40 |
41 | dflts = {[],[],[],[],0,0,0.2}; % define default values
42 |
43 | [nsamples,loglikelihood,priorpdf,prior_rnd,burnin,lastBurnin,beta] = ...
44 | internal.stats.parseArgs(pnames, dflts, varargin{:});
45 |
46 | %% Obtain N samples from the prior pdf f(T)
47 | j = 0; % Initialise loop for the transitional likelihood
48 | thetaj = prior_rnd(nsamples); % theta0 = N x D
49 | pj = 0; % p0 = 0 (initial tempering parameter)
50 | Dimensions = size(thetaj, 2); % size of the vector theta
51 |
52 | count = 1; % Counter
53 | samps(:,:,count) = thetaj;
54 | beta_j(count) = pj;
55 |
56 | %% Initialization of matrices and vectors
57 | thetaj1 = zeros(nsamples, Dimensions);
58 |
59 | %% Main loop
60 | while pj < 1
61 | j = j+1;
62 |
63 | %% Calculate the tempering parameter p(j+1):
64 | for l = 1:nsamples
65 | log_fD_T_thetaj(l) = loglikelihood(thetaj(l,:));
66 | end
67 | if any(isinf(log_fD_T_thetaj))
68 | error('The prior distribution is too far from the true region');
69 | end
70 | pj1 = calculate_pj1(nsamples, log_fD_T_thetaj, pj);
71 | fprintf('TMCMC: Iteration j = %2d, pj1 = %f\n', j, pj1);
72 |
73 | %% Compute the plausibility weight for each sample wrt f_{j+1}
74 | fprintf('Computing the weights ...\n');
75 | a = (pj1-pj)*log_fD_T_thetaj;
76 | wj = exp(a);
77 | wj_norm = wj./sum(wj); % normalization of the weights
78 |
79 | %% Compute S(j) = E[w{j}] (eq 15)
80 | S(j) = mean(wj);
81 |
82 | %% Do the resampling step to obtain N samples from f_{j+1}(theta) and
83 | % then perform Metropolis-Hastings on each of these samples using as a
84 | % stationary PDF "fj1"
85 | % fj1 = @(t) fT(t).*log_fD_T(t).^pj1; % stationary PDF (eq 11) f_{j+1}(theta)
86 | log_posterior = @(t) log(priorpdf(t)) + pj1*loglikelihood(t);
87 |
88 |
89 | % and using as proposal PDF a Gaussian centered at thetaj(idx,:) and
90 | % with covariance matrix equal to an scaled version of the covariance
91 | % matrix of fj1:
92 |
93 | % weighted mean
94 | mu = zeros(1, Dimensions);
95 | for l = 1:nsamples
96 | mu = mu + wj_norm(l)*thetaj(l,:); % 1 x N
97 | end
98 |
99 | % scaled covariance matrix of fj1 (eq 17)
100 | cov_gauss = zeros(Dimensions);
101 | for k = 1:nsamples
102 | % this formula is slightly different to eq 17 (the transpose)
103 | % because of the size of the vectors)m and because Ching and Chen
104 | % forgot to normalize the weight wj:
105 | tk_mu = thetaj(k,:) - mu;
106 | cov_gauss = cov_gauss + wj_norm(k)*(tk_mu'*tk_mu);
107 | end
108 | cov_gauss = beta^2 * cov_gauss;
109 | assert(~isinf(cond(cov_gauss)),'Something is wrong with the likelihood.')
110 |
111 | % Define the Proposal distribution:
112 | proppdf = @(x,y) prop_pdf(x, y, cov_gauss, priorpdf); %q(x,y) = q(x|y).
113 | proprnd = @(x) prop_rnd(x, cov_gauss, priorpdf);
114 |
115 | %% During the last iteration we require to do a better burnin in order
116 | % to guarantee the quality of the samples:
117 | if pj1 == 1
118 | burnin = lastBurnin;
119 | end
120 |
121 | %% Start N different Markov chains
122 | fprintf('Markov chains ...\n\n');
123 | idx = randsample(nsamples, nsamples, true, wj_norm);
124 |
125 | for i = 1:nsamples % For parallel, type: parfor
126 | %% Sample one point with probability wj_norm
127 |
128 | % smpl = mhsample(start, nsamples,
129 | % 'pdf', pdf, 'proppdf', proppdf, 'proprnd', proprnd);
130 | % start = row vector containing the start value of the Markov Chain,
131 | % nsamples = number of samples to be generated
132 | [thetaj1(i,:), acceptance_rate(i)] = mhsample(thetaj(idx(i), :), 1, ...
133 | 'logpdf', log_posterior, ...
134 | 'proppdf', proppdf, ...
135 | 'proprnd', proprnd, ...
136 | 'thin', 3, ...
137 | 'burnin', burnin);
138 | % According to Cheung and Beck (2009) - Bayesian model updating ...,
139 | % the initial samples from reweighting and the resample of samples of
140 | % fj, in general, do not exactly follow fj1, so that the Markov
141 | % chains must "burn-in" before samples follow fj1, requiring a large
142 | % amount of samples to be generated for each level.
143 |
144 | %% Adjust the acceptance rate (optimal = 23%)
145 | % See: http://www.dms.umontreal.ca/~bedard/Beyond_234.pdf
146 | %{
147 | if acceptance_rate < 0.3
148 | % Many rejections means an inefficient chain (wasted computation
149 | %time), decrease the variance
150 | beta = 0.99*beta;
151 | elseif acceptance_rate > 0.5
152 | % High acceptance rate: Proposed jumps are very close to current
153 | % location, increase the variance
154 | beta = 1.01*beta;
155 | end
156 | %}
157 | end
158 | fprintf('\n');
159 | acceptance(count) = mean(acceptance_rate);
160 | %% Prepare for the next iteration
161 | count = count+1;
162 | samps(:,:,count) = thetaj1;
163 | thetaj = thetaj1;
164 | pj = pj1;
165 | beta_j(count) = pj;
166 | end
167 |
168 | % estimation of f(D) -- this is the normalization constant in Bayes
169 | log_fD = sum(log(S(1:j)));
170 |
171 | %% Description of outputs:
172 |
173 | output.allsamples = samps; % To show samples from all transitional distributions
174 | output.samples = samps(:,:,end); % To only show samples from the final posterior
175 | output.log_evidence = log_fD; % To generate the logarithmic of the evidence
176 | output.acceptance = acceptance; % To show the mean acceptance rates for all iterations
177 | output.beta = beta_j; % To show the values of temepring parameters, beta_j
178 |
179 | return; % End
180 |
181 |
182 | %% Calculate the tempering parameter p(j+1)
183 | function pj1 = calculate_pj1(nsamples, log_fD_T_thetaj, pj)
184 | % find pj1 such that the Effective Sample Size (ESS) equals N/2:
185 | %
186 | % 1
187 | % ------------------ >= threshold
188 | % sum[(hat_wj)^2]
189 | %
190 | % here:
191 | % hat_wj = wj/sum(wj)
192 | % size(thetaj) = N x D,
193 | % wj = fD_T(thetaj).^(pj1 - pj)
194 | % e = pj1 - pj
195 |
196 | threshold = nsamples;
197 |
198 | % Note the following trick in order to calculate e:
199 | % Take into account that e>=0
200 | wj = @(e) exp(abs(e)*log_fD_T_thetaj); % N x 1
201 | fmin = @(e) (sum(wj(e)))^2 - (threshold .* sum((wj(e)).^2)) + realmin;
202 | e = abs(fzero(fmin, 0)); % e is >= 0, and fmin is an even function
203 | if isnan(e)
204 | error('There is an error finding e');
205 | end
206 |
207 | pj1 = min(1, pj + e);
208 |
209 | return; % End
210 |
211 | function proppdf = prop_pdf(x, mu, covmat, box)
212 | % This is the Proposal PDF for the Markov Chain.
213 |
214 | % Box function is the Prior PDF in the feasible region.
215 | % So if a point is out of bounds, this function will
216 | % return 0.
217 |
218 | proppdf = mvnpdf(x, mu, covmat).*box(x); %q(x,y) = q(x|y).
219 |
220 | return;
221 |
222 |
223 | function proprnd = prop_rnd(mu, covmat, box)
224 | % Sampling from the proposal PDF for the Markov Chain.
225 |
226 | while true
227 | proprnd = mvnrnd(mu, covmat, 1);
228 | if box(proprnd)
229 | % The box function is the Prior PDF in the feasible region.
230 | % If a point is out of bounds, this function will return 0 = false.
231 | break;
232 | end
233 | end
234 |
235 | return
236 |
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/angleCalc.m:
--------------------------------------------------------------------------------
1 | function theta = angleCalc(S,C,out_mode)
2 | %% This function computes the angle from sin and cos values (-180,180] degree.
3 | %
4 | % Usage:
5 | % theta = angleCalc(S,C,out_mode)
6 | %
7 | % Input:
8 | % S: Sine value of the angle
9 | % C: Cosine value of the angle
10 | % out_mode: 'deg' OR 'rad'
11 | % Note: default output mode is in degree
12 | %
13 | % Output:
14 | % theta: Angles in degrees or radians.
15 | %
16 | % Example:
17 | % theta = angleCalc(sin(-2*pi/3),cos(-2*pi/3))
18 | % theta = -120;
19 | % theta= angleCalc(sin(2*pi/3),cos(2*pi/3),'rad')
20 | % theta= 2.0944 [rad]
21 | % --------------Disi A Jun 25, 2013
22 | %% Define the function:
23 | if nargin < 3
24 | out_mode='deg';
25 | end
26 |
27 | if strcmp(out_mode,'deg')
28 | cons = 180/pi;
29 | else
30 | cons = 1;
31 | end
32 |
33 | for i = 1:length(S)
34 | theta(i) = asin(S(i));
35 | if C(i) < 0
36 | if S(i) > 0
37 | theta(i) = pi - theta(i);
38 | elseif S(i) < 0
39 | theta(i) = - pi - theta(i);
40 | else % If S(i) = 0
41 | theta(i) = theta(i) + pi;
42 | end
43 | end
44 |
45 | theta(i) = theta(i) .* cons;
46 | end
47 | end
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/areaMe.m:
--------------------------------------------------------------------------------
1 | function [outputArg1] = areaMe(D1,D2)
2 | %AREAME Computes the area between two ECDFs
3 | % It does not work with a single datum.
4 | %
5 | % .
6 | % . by The Liverpool Git Pushers
7 | if length(D1)>length(D2)
8 | d1 = D2(:);
9 | d2 = D1(:);
10 | else
11 | d1 = D1(:);
12 | d2 = D2(:);
13 | end
14 | [Pxs,xs] = ecdf_Lpool(d1); % Compute the ecdf of the data sets
15 | [Pys,ys] = ecdf_Lpool(d2);
16 | Pys_eqx = Pxs;
17 | Pys_pure = Pys(2:end-1); % this does not work with a single datum
18 | Pall = sort([Pys_eqx;Pys_pure]);
19 | ys_eq_all = zeros(length(Pall),1);
20 | ys_eq_all(1)=ys(1);
21 | ys_eq_all(end)=ys(end);
22 | for k=2:length(Pall)-1
23 | ys_eq_all(k,1) = interpCDF_2(ys,Pys,Pall(k));
24 | end
25 | xs_eq_all = zeros(length(Pall),1);
26 | xs_eq_all(1)=xs(1);
27 | xs_eq_all(end)=xs(end);
28 | for k=2:length(Pall)-1
29 | xs_eq_all(k,1) = interpCDF_2(xs,Pxs,Pall(k));
30 | end
31 | diff_all_s = abs(ys_eq_all-xs_eq_all);
32 | diff_all_s = diff_all_s(2:end);
33 | diff_all_p = diff(Pall);
34 | area = diff_all_s' * diff_all_p;
35 | outputArg1 = area;
36 | end
37 |
38 |
39 | function [outputArg1] = interpCDF_2(xd,yd,pvalue)
40 | %INTERPCDF Summary of this function goes here
41 | % Detailed explanation goes here
42 | %
43 | % .
44 | % . by The Liverpool Git Pushers
45 |
46 | % [yd,xd]=ecdf_Lpool(data);
47 | beforr = diff(pvalue <= yd) == 1; % && diff(0.5>pv) == -1;
48 | beforrr = [0;beforr(:)];
49 | if pvalue==0
50 | xvalue = xd(1);
51 | else
52 | xvalue = xd(beforrr==1);
53 | end
54 | outputArg1 = xvalue;
55 | end
56 |
57 |
58 | function [ps,xs] = ecdf_Lpool(x)
59 |
60 | xs = sort(x);
61 | xs = [xs(1);xs(:)];
62 | ps = linspace(0,1,length(xs))';
63 |
64 | end
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/blackbox_model.m:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/Adolphus8/Bayesian-Model-Updating-Tutorials/f88cbbc1bac5e4caab6f5fd620d76a25c50defa5/Alternative_TMCMC_Transition_Criteria/blackbox_model.m
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/example_SDOF_System_Coulomb_Friction_numerical.m:
--------------------------------------------------------------------------------
1 | %% Numerical Example: SDOF System with Coulomb Friction
2 | %
3 | % This set-up is based on the Single Degree-of-Freedom Dynamical System
4 | % subjected to Coulomb Friction Force that is presented in the literature:
5 | %
6 | % L. Marino and A. Cicirello (2020). Experimental investigation of a single-
7 | % degree-of-freedom system with Coulomb friction. Nonlinear Dynamics, 99(3),
8 | % 1781-1799. doi: 10.1007/s11071-019-05443-2
9 | %
10 | % This set-up is applicable only for the Base motion (with fixed wall) case.
11 | %-------------------------------------------------------------------------%
12 | %% Load Numerical Data:
13 | clc; clear;
14 | load('noisy_data');
15 |
16 | %% Define key parameters:
17 | omega_n = 3.086; % Natural frequency of the structure [Hz]
18 | driving_force = 2.5; % Driving force amplitude by the rotor [N]
19 | r_nom = data.frequency_ratio; % Nominal values of the dimensionless input frequency ratios
20 | coulomb_force = flip(driving_force .* data.force_ratio); % Nominal values of the Coulomb Frictions [N]
21 |
22 | %% Define the data-set:
23 | Ndata = 15; % Data size per time-step
24 | phase_angle_analytical = zeros(Ndata,1);
25 | phase_angle_analytical(:,1) = data.phase_angle.f010;
26 | phase_angle_analytical(:,2) = data.phase_angle.f025;
27 | phase_angle_analytical(:,3) = data.phase_angle.f040;
28 | phase_angle_analytical(:,4) = data.phase_angle.f055;
29 |
30 | % Consolidate the "noisy" Phase angle data:
31 | sigma_phi = 2; % True value of measurement noise for Phase angles
32 | % phase_angle_noisy = phase_angle_analytical + sigma_phi.*randn(Ndata, length(data.force_ratio));
33 | phase_angle_noisy = data.noisy_phase_angles;
34 | data_phase_angle = phase_angle_noisy(2:11,:); % Take only 10 data for Bayesian model updating
35 |
36 | % Consolidate the "noisy" input Frequency data:
37 | sigma_r = 0.01; % True value of measurement noise for Frequency ratio
38 | r_v = data.noisy_frequency_ratios;
39 | frequency_data = r_v .* omega_n;
40 | data_frequency = frequency_data(2:11,:); % Take only 10 data for Bayesian model updating
41 |
42 | % Generate Analytical solution through Den-Hartog's solution:
43 | beta_nom = data.force_ratio; % Nominal force ratio
44 | output = DenHartogHarmonic(beta_nom');
45 | r_an = output.frequency_ratios; % The output frequency ratios
46 | phase_an = output.phase_angles; % The output analytical phase angles
47 | phase_bound = output.phase_bound; % The phase angle bound defined by Den-Hartog's Boundary
48 |
49 | %% To plot the Phase angles vs Frequency ratio curves:
50 | colors = [0 0 1; 0 0.5 0; 1 0 0; 1 0 1];
51 |
52 | figure;
53 | hold on; box on; grid on;
54 | for ib = 3 % To plot for different Friction Force ratio
55 | % Numerical scatterplot:
56 | plot([r_v(:,ib)], phase_angle_noisy(:,ib), 'o', 'color', colors(ib,:), 'linewidth', 2);
57 | % Analytical plot:
58 | plot([r_an], phase_an(ib+1,:), '--', 'color', colors(ib,:), 'linewidth', 1);
59 | end
60 | plot([r_an], phase_bound,'-- k');
61 | legend(['Data for F_{\mu} = ',num2str(1.0864, '%.3f'), ' N'],...
62 | 'Analytical solution', 'Den-Hartog''s Boundary', 'linewidth', 2, 'location', 'southeast');
63 | xlabel('$r$', 'Interpreter', 'latex'); ylabel('$\phi$ $[deg]$', 'Interpreter', 'latex');
64 | xlim([0, 2]); ylim([0 180]); set(gca,'FontSize',20);
65 |
66 | %% Bayesian Model Updating Set-up:
67 | % The epistemic parameters to be inferred are the following:
68 | % {Coulomb Force, Natural Frequency, Frequency Ratio Noise, Phase Angle Noise}
69 |
70 | % Define the Prior distribution:
71 | lowerbound = [0.01, 0.001, 0.001]; upperbound = [10, 10, 1];
72 | prior_coulomb = @(x) unifpdf(x, lowerbound(1), upperbound(1)); % Prior for Coulomb Friction
73 | prior_omega = @(x) unifpdf(x, lowerbound(2), upperbound(2)); % Prior for Natural Frequency
74 | prior_sigma_phi = @(x) unifpdf(x, lowerbound(2), upperbound(2)); % Prior for Phase Angle Noise
75 | prior_sigma_r = @(x) unifpdf(x, lowerbound(3), upperbound(3)); % Prior for Frequency Ratio Noise
76 |
77 | prior_pdf = @(x) prior_coulomb(x(:,1)) .* prior_omega(x(:,2)) .* ...
78 | prior_sigma_phi(x(:,3)) .* prior_sigma_r(x(:,4));
79 | prior_rnd = @(N) [unifrnd(lowerbound(1), upperbound(1), N, 1), ...
80 | unifrnd(lowerbound(2), upperbound(2), N, 1), ...
81 | unifrnd(lowerbound(2), upperbound(2), N, 1), ...
82 | unifrnd(lowerbound(3), upperbound(3), N, 1)];
83 |
84 | % Define the loglikelihood function:
85 | model = @(x,f) blackbox_model(x, f, driving_force);
86 | t = 3;
87 | logL = @(x) loglikelihood(x, model, data_phase_angle(:,t), data_frequency(:,t), r_v(2:11), r_nom(2:11));
88 |
89 | %% Define Bayesian Model Updating Parameters:
90 |
91 | Nsamples = 1000; % No. of samples to generate from the Posterior
92 | Nbatch = 1; % No. of sample runs to perform
93 | Ncores = 12;
94 | TMCMC1 = cell(Nbatch,1); TMCMC2 = cell(Nbatch,1);
95 | timeTMCMC1 = zeros(Nbatch,1); timeTMCMC2 = zeros(Nbatch,1);
96 |
97 | %% Perform Bayesian Model Updating via TMCMC and TMCMC-II:
98 |
99 | % Initiate the samplers:
100 |
101 | parpool(Ncores)
102 | parfor r = 1:Nbatch
103 |
104 | fprintf('Batch no.: %d \n',r)
105 |
106 | tic;
107 | TMCMC1{r,1} = TMCMCsampler('nsamples',Nsamples,'loglikelihood',logL,'priorpdf',prior_pdf,...
108 | 'priorrnd',prior_rnd,'burnin',0,'lastburnin',0);
109 | timeTMCMC1(r,1) = toc;
110 | fprintf('Time elapsed is for the TMCMC sampler: %f \n',timeTMCMC1(r,1))
111 |
112 | tic;
113 | TMCMC2{r,1} = TMCMCsampler2('nsamples',Nsamples,'loglikelihood',logL,'priorpdf',prior_pdf,...
114 | 'priorrnd',prior_rnd,'burnin',0,'lastburnin',0);
115 | timeTMCMC2(r,1) = toc;
116 | fprintf('Time elapsed is for the TMCMC-II sampler: %f \n',timeTMCMC2(r,1))
117 | end
118 | %% Save the data:
119 | save('ESREL2023')
120 |
121 | %%
122 | %{
123 | TMCMC1_cell = cell(50,1); TMCMC2_cell = cell(50,1);
124 | timeTMCMC1_vec = zeros(50,1); timeTMCMC2_vec = zeros(50,1);
125 |
126 | for i = 1:22
127 | load('ESREL2023_1.mat', 'TMCMC1', 'TMCMC2', 'timeTMCMC1', 'timeTMCMC2')
128 | TMCMC1_cell{i} = TMCMC1{i}; TMCMC2_cell{i} = TMCMC2{i};
129 | timeTMCMC1_vec(i) = timeTMCMC1(i); timeTMCMC2_vec(i) = timeTMCMC2(i);
130 | end
131 |
132 | for i = 23:30
133 | load('ESREL2023_2.mat', 'TMCMC1', 'TMCMC2', 'timeTMCMC1', 'timeTMCMC2')
134 | TMCMC1_cell{i} = TMCMC1{i-22}; TMCMC2_cell{i} = TMCMC2{i-22};
135 | timeTMCMC1_vec(i) = timeTMCMC1(i-22); timeTMCMC2_vec(i) = timeTMCMC2(i-22);
136 | end
137 |
138 | for i = 31:40
139 | load('ESREL2023_3.mat', 'TMCMC1', 'TMCMC2', 'timeTMCMC1', 'timeTMCMC2')
140 | TMCMC1_cell{i} = TMCMC1{i-30}; TMCMC2_cell{i} = TMCMC2{i-30};
141 | timeTMCMC1_vec(i) = timeTMCMC1(i-30); timeTMCMC2_vec(i) = timeTMCMC2(i-30);
142 | end
143 |
144 | for i = 41:50
145 | load('ESREL2023_4.mat', 'TMCMC1', 'TMCMC2', 'timeTMCMC1', 'timeTMCMC2')
146 | TMCMC1_cell{i} = TMCMC1{i-40}; TMCMC2_cell{i} = TMCMC2{i-40};
147 | timeTMCMC1_vec(i) = timeTMCMC1(i-40); timeTMCMC2_vec(i) = timeTMCMC2(i-40);
148 | end
149 |
150 | TMCMC1 = TMCMC1_cell; TMCMC2 = TMCMC2_cell; timeTMCMC1 = timeTMCMC1_vec; timeTMCMC2 = timeTMCMC2_vec;
151 | save('ESREL2023.mat', 'TMCMC1', 'TMCMC2', 'timeTMCMC1', 'timeTMCMC2')
152 | %}
153 |
154 | %% P-box Analysis:
155 | load('ESREL2023')
156 |
157 | sampsTMCMC1 = zeros(Nsamples, length(TMCMC1),2); sampsTMCMC2 = zeros(Nsamples, length(TMCMC2),2);
158 | pboxTMCMC1 = zeros(Nsamples,2,2); pboxTMCMC2 = zeros(Nsamples,2,2);
159 | for i = 1:length(TMCMC1)
160 | cell1 = TMCMC1{i}; cell2 = TMCMC2{i};
161 | samps1 = cell1.samples; samps2 = cell2.samples;
162 | sampsTMCMC1(:,i,1) = sort(samps1(:,1), 'ascend'); sampsTMCMC1(:,i,2) = sort(samps1(:,2), 'ascend');
163 | sampsTMCMC2(:,i,1) = sort(samps2(:,1), 'ascend'); sampsTMCMC2(:,i,2) = sort(samps2(:,2), 'ascend');
164 | end
165 |
166 | for j = 1:Nsamples
167 | pboxTMCMC1(j,:,1) = [min(sampsTMCMC1(j,:,1)),max(sampsTMCMC1(j,:,1))]; pboxTMCMC1(j,:,2) = [min(sampsTMCMC1(j,:,2)),max(sampsTMCMC1(j,:,2))];
168 | pboxTMCMC2(j,:,1) = [min(sampsTMCMC2(j,:,1)),max(sampsTMCMC2(j,:,1))]; pboxTMCMC2(j,:,2) = [min(sampsTMCMC2(j,:,2)),max(sampsTMCMC2(j,:,2))];
169 | end
170 |
171 | % Pbox of estimates:
172 | figure;
173 | subplot(1,2,1)
174 | hold on; box on; grid on;
175 | [f1,x1] = ecdf(pboxTMCMC1(:,1,1)); [f2,x2] = ecdf(pboxTMCMC1(:,2,1));
176 | stairs(x1,f1, 'b', 'linewidth', 2); stairs(x2,f2, 'b', 'linewidth', 2, 'handlevisibility', 'off');
177 | plot([min(x1),min(x2)],[0,0], 'b', 'linewidth', 2, 'handlevisibility', 'off');
178 | plot([max(x1),max(x2)],[1,1], 'b', 'linewidth', 2, 'handlevisibility', 'off');
179 |
180 | [f1,x1] = ecdf(pboxTMCMC2(:,1,1)); [f2,x2] = ecdf(pboxTMCMC2(:,2,1));
181 | stairs(x1,f1, 'r', 'linewidth', 2); stairs(x2,f2, 'r', 'linewidth', 2, 'handlevisibility', 'off');
182 | plot([min(x1),min(x2)],[0,0], 'r', 'linewidth', 2, 'handlevisibility', 'off');
183 | plot([max(x1),max(x2)],[1,1], 'r', 'linewidth', 2, 'handlevisibility', 'off');
184 | xline(1.0864, 'k--', 'linewidth', 2);
185 | legend('P-box TMCMC', 'P-box TMCMC-II', 'True value F_{\mu} = 1.086 [N]', 'linewidth', 2)
186 | xlabel('$F_{\mu}$ $[N]$', 'Interpreter', 'latex'); ylabel('ECDF value'); set(gca, 'Fontsize', 18)
187 | xlim([0.7, 1.8])
188 |
189 | subplot(1,2,2)
190 | hold on; box on; grid on;
191 | [f1,x1] = ecdf(pboxTMCMC1(:,1,2)); [f2,x2] = ecdf(pboxTMCMC1(:,2,2));
192 | stairs(x1,f1, 'b', 'linewidth', 2); stairs(x2,f2, 'b', 'linewidth', 2, 'handlevisibility', 'off');
193 | plot([min(x1),min(x2)],[0,0], 'b', 'linewidth', 2, 'handlevisibility', 'off');
194 | plot([max(x1),max(x2)],[1,1], 'b', 'linewidth', 2, 'handlevisibility', 'off');
195 |
196 | [f1,x1] = ecdf(pboxTMCMC2(:,1,2)); [f2,x2] = ecdf(pboxTMCMC2(:,2,2));
197 | stairs(x1,f1, 'r', 'linewidth', 2); stairs(x2,f2, 'r', 'linewidth', 2, 'handlevisibility', 'off');
198 | plot([min(x1),min(x2)],[0,0], 'r', 'linewidth', 2, 'handlevisibility', 'off');
199 | plot([max(x1),max(x2)],[1,1], 'r', 'linewidth', 2, 'handlevisibility', 'off');
200 | xline(3.086, 'k--', 'linewidth', 2);
201 | legend('P-box TMCMC', 'P-box TMCMC-II', 'True value \omega_n = 3.086 [Hz]', 'linewidth', 2)
202 | xlabel('$\omega_n$ $[Hz]$', 'Interpreter', 'latex'); ylabel('ECDF value'); set(gca, 'Fontsize', 18)
203 |
204 | area_mat = zeros(2,2);
205 | area_mat(1,1) = areaMe(pboxTMCMC1(:,1,1),pboxTMCMC1(:,2,1)); area_mat(1,2) = areaMe(pboxTMCMC1(:,1,2),pboxTMCMC1(:,2,2));
206 | area_mat(2,1) = areaMe(pboxTMCMC2(:,1,1),pboxTMCMC2(:,2,1)); area_mat(2,2) = areaMe(pboxTMCMC2(:,1,2),pboxTMCMC2(:,2,2));
207 | T = array2table(area_mat,'VariableNames', ...
208 | {'Coulomb_Friction_Pbox_area', 'Natural_Frequency_Pbox_area'},...
209 | 'RowNames', {'TMCMC', 'TMCMC-II'});
210 |
211 | area_vec = [0.081508, 0.17135; 0.008878, 0.0089148];
212 | x = categorical({'F_{\mu} [N]', '\omega_{n} [Hz]'}); x = reordercats(x,{'F_{\mu} [N]', '\omega_{n} [Hz]'});
213 |
214 | figure;
215 | hold on; box on; grid on;
216 | b = bar(x, area_vec);
217 | xlabel('Parameter'); ylabel('Area of P-box');
218 | set(gca, 'Fontsize', 18); legend('TMCMC', 'TMCMC-II', 'linewidth', 2)
219 |
220 | %% Mean Analysis:
221 |
222 | meanTMCMC1 = zeros(length(TMCMC1),2); meanTMCMC2 = zeros(length(TMCMC2),2);
223 | betaTMCMC1 = zeros(length(TMCMC1),1); betaTMCMC2 = zeros(length(TMCMC2),1);
224 |
225 | for i = 1:length(TMCMC1)
226 | cell1 = TMCMC1{i}; cell2 = TMCMC2{i};
227 | samps1 = cell1.samples; samps2 = cell2.samples;
228 | meanTMCMC1(i,:) = mean(samps1(:,1:2)); meanTMCMC2(i,:) = mean(samps2(:,1:2));
229 | betaTMCMC1(i) = length(cell1.beta)-1; betaTMCMC2(i) = length(cell2.beta)-1;
230 | end
231 |
232 | figure;
233 | subplot(2,2,1)
234 | hold on; box on; grid on;
235 | histogram(meanTMCMC1(:,1))
236 | xlabel('$F_{\mu}$ $[N]$', 'Interpreter', 'latex'); ylabel('Count'); set(gca, 'Fontsize', 18)
237 | title('Mean values TMCMC')
238 | subplot(2,2,2)
239 | hold on; box on; grid on;
240 | histogram(meanTMCMC1(:,2))
241 | xlabel('$\omega_n$ $[Hz]$', 'Interpreter', 'latex'); ylabel('Count'); set(gca, 'Fontsize', 18)
242 | title('Mean values TMCMC')
243 |
244 | subplot(2,2,3)
245 | hold on; box on; grid on;
246 | histogram(meanTMCMC2(:,1))
247 | xlabel('$F_{\mu}$ $[N]$', 'Interpreter', 'latex'); ylabel('Count'); set(gca, 'Fontsize', 18)
248 | title('Mean values TMCMC2')
249 | subplot(2,2,4)
250 | hold on; box on; grid on;
251 | histogram(meanTMCMC2(:,2))
252 | xlabel('$\omega_n$ $[Hz]$', 'Interpreter', 'latex'); ylabel('Count'); set(gca, 'Fontsize', 18)
253 | title('Mean values TMCMC2')
254 |
255 | figure;
256 | hold on; box on; grid on;
257 | [f1,x1] = ecdf(betaTMCMC1); [f2,x2] = ecdf(betaTMCMC2);
258 | stairs(x1,f1, 'b', 'linewidth', 2); stairs(x2,f2, 'r', 'linewidth', 2);
259 | plot([11,12], [1,1], 'b', 'linewidth',2, 'handlevisibility', 'off')
260 | xticks([10:12]); xlabel('No. of Transition steps'); ylabel('ECDF values'); set(gca, 'Fontsize', 18)
261 |
262 | beta_vec = [6,5 ; 44,36 ; 0,9];
263 | iterations = [9, 10, 11];
264 |
265 | figure;
266 | hold on; box on; grid on;
267 | b = bar(iterations, beta_vec);
268 | xlabel('No. of Iterations'); ylabel('Count'); xticks([9:11])
269 | set(gca, 'Fontsize', 18); legend('TMCMC', 'TMCMC-II', 'linewidth', 2)
270 |
271 |
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/loglikelihood.m:
--------------------------------------------------------------------------------
1 | function logL = loglikelihood(theta, blackbox_model, phase_angle_data, frequency_data, frequency_ratio_exp, frequency_ratio_nom)
2 | %% Function-handle of the Loglikelihood function:
3 | % This function-handle computes the log-likelihood values.
4 | %-------------------------------------------------------------------------%
5 | %
6 | % Inputs:
7 | % theta: N x 4 input matrix of the epistemic parameters whereby:
8 | % - theta(:,1): First dimension is that of the time-varying Coulomb Friction [N];
9 | % - theta(:,2): Second dimension is that of the static Natural Frequency [Hz];
10 | % - theta(:,3): Third dimension is that of the static noise for Phase Angle measurements [deg];
11 | % - theta(:,4): Fourth dimension is that of the static noise for Frequency ratio measurements;
12 | % Note: N is the sample size, the number of theta to generate from the posterior.
13 | %
14 | % blackbox_model: Blackbox function-handle (function of theta) used for model evaluation;
15 | % phase_angle_data: N_e x 1 input vector of the phase angles measured from the experiment [deg];
16 | % frequency_data: N_e x 1 input vector of the driving frequencies used for the experiment [rad/s];
17 | % frequency_ratio_nom: N_e x 1 input vector of nominal frequency ratio;
18 | % frequency_ratio_exp: N_e x 1 input vector of experimental frequency ratio;
19 | % Note: N_e is the number of experimental data taken.
20 | %
21 | % Output:
22 | % logL: N x 1 vector of loglikelihood output values;
23 | %
24 | %-------------------------------------------------------------------------%
25 | %% Define the Function-handle:
26 |
27 | % Initiate the empty vector of logL:
28 | logL = zeros(size(theta,1),1);
29 |
30 | for i = 1:size(theta,1)
31 | %% Generate the model output:
32 | model_output = blackbox_model(theta(i,1), frequency_ratio_exp);
33 |
34 | % Generate N_e x 1 model output of the Phase angles:
35 | phase_angle_model = model_output.phase_angles;
36 |
37 | %% Compute the loglikelihood for Phase Angles:
38 |
39 | logL_phi = - 0.5 .* (1./(theta(i,3)).^2) .* sum((phase_angle_data - phase_angle_model).^2) - ...
40 | size(phase_angle_data,1).*log(sqrt(2*pi).*theta(i,3));
41 |
42 | %% Compute the loglikelihood for Frequency Ratios:
43 |
44 | % Generate N_e x 1 model output of the frequency ratios:
45 | freq_ratio_model = frequency_data./theta(i,2);
46 |
47 | logL_r = - 0.5 .* (1./(theta(i,4)).^2) .* sum((frequency_ratio_nom - freq_ratio_model).^2) - ...
48 | size(frequency_data,1).*log(sqrt(2*pi).*theta(i,4));
49 |
50 | %% Compute the overall loglikelihood value:
51 |
52 | logL(i) = logL_phi + logL_r;
53 |
54 | % Set logL(i) = -1e10 if logL is NaN or Inf:
55 | if isnan(logL(i)) || isinf(logL(i))
56 | logL(i) = -1e10;
57 | end
58 |
59 | end
60 | end
61 |
62 |
--------------------------------------------------------------------------------
/Alternative_TMCMC_Transition_Criteria/test:
--------------------------------------------------------------------------------
1 |
2 |
--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
1 | GNU GENERAL PUBLIC LICENSE
2 | Version 3, 29 June 2007
3 |
4 | Copyright (C) 2007 Free Software Foundation, Inc.
5 | Everyone is permitted to copy and distribute verbatim copies
6 | of this license document, but changing it is not allowed.
7 |
8 | Preamble
9 |
10 | The GNU General Public License is a free, copyleft license for
11 | software and other kinds of works.
12 |
13 | The licenses for most software and other practical works are designed
14 | to take away your freedom to share and change the works. By contrast,
15 | the GNU General Public License is intended to guarantee your freedom to
16 | share and change all versions of a program--to make sure it remains free
17 | software for all its users. We, the Free Software Foundation, use the
18 | GNU General Public License for most of our software; it applies also to
19 | any other work released this way by its authors. You can apply it to
20 | your programs, too.
21 |
22 | When we speak of free software, we are referring to freedom, not
23 | price. Our General Public Licenses are designed to make sure that you
24 | have the freedom to distribute copies of free software (and charge for
25 | them if you wish), that you receive source code or can get it if you
26 | want it, that you can change the software or use pieces of it in new
27 | free programs, and that you know you can do these things.
28 |
29 | To protect your rights, we need to prevent others from denying you
30 | these rights or asking you to surrender the rights. Therefore, you have
31 | certain responsibilities if you distribute copies of the software, or if
32 | you modify it: responsibilities to respect the freedom of others.
33 |
34 | For example, if you distribute copies of such a program, whether
35 | gratis or for a fee, you must pass on to the recipients the same
36 | freedoms that you received. You must make sure that they, too, receive
37 | or can get the source code. And you must show them these terms so they
38 | know their rights.
39 |
40 | Developers that use the GNU GPL protect your rights with two steps:
41 | (1) assert copyright on the software, and (2) offer you this License
42 | giving you legal permission to copy, distribute and/or modify it.
43 |
44 | For the developers' and authors' protection, the GPL clearly explains
45 | that there is no warranty for this free software. For both users' and
46 | authors' sake, the GPL requires that modified versions be marked as
47 | changed, so that their problems will not be attributed erroneously to
48 | authors of previous versions.
49 |
50 | Some devices are designed to deny users access to install or run
51 | modified versions of the software inside them, although the manufacturer
52 | can do so. This is fundamentally incompatible with the aim of
53 | protecting users' freedom to change the software. The systematic
54 | pattern of such abuse occurs in the area of products for individuals to
55 | use, which is precisely where it is most unacceptable. Therefore, we
56 | have designed this version of the GPL to prohibit the practice for those
57 | products. If such problems arise substantially in other domains, we
58 | stand ready to extend this provision to those domains in future versions
59 | of the GPL, as needed to protect the freedom of users.
60 |
61 | Finally, every program is threatened constantly by software patents.
62 | States should not allow patents to restrict development and use of
63 | software on general-purpose computers, but in those that do, we wish to
64 | avoid the special danger that patents applied to a free program could
65 | make it effectively proprietary. To prevent this, the GPL assures that
66 | patents cannot be used to render the program non-free.
67 |
68 | The precise terms and conditions for copying, distribution and
69 | modification follow.
70 |
71 | TERMS AND CONDITIONS
72 |
73 | 0. Definitions.
74 |
75 | "This License" refers to version 3 of the GNU General Public License.
76 |
77 | "Copyright" also means copyright-like laws that apply to other kinds of
78 | works, such as semiconductor masks.
79 |
80 | "The Program" refers to any copyrightable work licensed under this
81 | License. Each licensee is addressed as "you". "Licensees" and
82 | "recipients" may be individuals or organizations.
83 |
84 | To "modify" a work means to copy from or adapt all or part of the work
85 | in a fashion requiring copyright permission, other than the making of an
86 | exact copy. The resulting work is called a "modified version" of the
87 | earlier work or a work "based on" the earlier work.
88 |
89 | A "covered work" means either the unmodified Program or a work based
90 | on the Program.
91 |
92 | To "propagate" a work means to do anything with it that, without
93 | permission, would make you directly or secondarily liable for
94 | infringement under applicable copyright law, except executing it on a
95 | computer or modifying a private copy. Propagation includes copying,
96 | distribution (with or without modification), making available to the
97 | public, and in some countries other activities as well.
98 |
99 | To "convey" a work means any kind of propagation that enables other
100 | parties to make or receive copies. Mere interaction with a user through
101 | a computer network, with no transfer of a copy, is not conveying.
102 |
103 | An interactive user interface displays "Appropriate Legal Notices"
104 | to the extent that it includes a convenient and prominently visible
105 | feature that (1) displays an appropriate copyright notice, and (2)
106 | tells the user that there is no warranty for the work (except to the
107 | extent that warranties are provided), that licensees may convey the
108 | work under this License, and how to view a copy of this License. If
109 | the interface presents a list of user commands or options, such as a
110 | menu, a prominent item in the list meets this criterion.
111 |
112 | 1. Source Code.
113 |
114 | The "source code" for a work means the preferred form of the work
115 | for making modifications to it. "Object code" means any non-source
116 | form of a work.
117 |
118 | A "Standard Interface" means an interface that either is an official
119 | standard defined by a recognized standards body, or, in the case of
120 | interfaces specified for a particular programming language, one that
121 | is widely used among developers working in that language.
122 |
123 | The "System Libraries" of an executable work include anything, other
124 | than the work as a whole, that (a) is included in the normal form of
125 | packaging a Major Component, but which is not part of that Major
126 | Component, and (b) serves only to enable use of the work with that
127 | Major Component, or to implement a Standard Interface for which an
128 | implementation is available to the public in source code form. A
129 | "Major Component", in this context, means a major essential component
130 | (kernel, window system, and so on) of the specific operating system
131 | (if any) on which the executable work runs, or a compiler used to
132 | produce the work, or an object code interpreter used to run it.
133 |
134 | The "Corresponding Source" for a work in object code form means all
135 | the source code needed to generate, install, and (for an executable
136 | work) run the object code and to modify the work, including scripts to
137 | control those activities. However, it does not include the work's
138 | System Libraries, or general-purpose tools or generally available free
139 | programs which are used unmodified in performing those activities but
140 | which are not part of the work. For example, Corresponding Source
141 | includes interface definition files associated with source files for
142 | the work, and the source code for shared libraries and dynamically
143 | linked subprograms that the work is specifically designed to require,
144 | such as by intimate data communication or control flow between those
145 | subprograms and other parts of the work.
146 |
147 | The Corresponding Source need not include anything that users
148 | can regenerate automatically from other parts of the Corresponding
149 | Source.
150 |
151 | The Corresponding Source for a work in source code form is that
152 | same work.
153 |
154 | 2. Basic Permissions.
155 |
156 | All rights granted under this License are granted for the term of
157 | copyright on the Program, and are irrevocable provided the stated
158 | conditions are met. This License explicitly affirms your unlimited
159 | permission to run the unmodified Program. The output from running a
160 | covered work is covered by this License only if the output, given its
161 | content, constitutes a covered work. This License acknowledges your
162 | rights of fair use or other equivalent, as provided by copyright law.
163 |
164 | You may make, run and propagate covered works that you do not
165 | convey, without conditions so long as your license otherwise remains
166 | in force. You may convey covered works to others for the sole purpose
167 | of having them make modifications exclusively for you, or provide you
168 | with facilities for running those works, provided that you comply with
169 | the terms of this License in conveying all material for which you do
170 | not control copyright. Those thus making or running the covered works
171 | for you must do so exclusively on your behalf, under your direction
172 | and control, on terms that prohibit them from making any copies of
173 | your copyrighted material outside their relationship with you.
174 |
175 | Conveying under any other circumstances is permitted solely under
176 | the conditions stated below. Sublicensing is not allowed; section 10
177 | makes it unnecessary.
178 |
179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
180 |
181 | No covered work shall be deemed part of an effective technological
182 | measure under any applicable law fulfilling obligations under article
183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or
184 | similar laws prohibiting or restricting circumvention of such
185 | measures.
186 |
187 | When you convey a covered work, you waive any legal power to forbid
188 | circumvention of technological measures to the extent such circumvention
189 | is effected by exercising rights under this License with respect to
190 | the covered work, and you disclaim any intention to limit operation or
191 | modification of the work as a means of enforcing, against the work's
192 | users, your or third parties' legal rights to forbid circumvention of
193 | technological measures.
194 |
195 | 4. Conveying Verbatim Copies.
196 |
197 | You may convey verbatim copies of the Program's source code as you
198 | receive it, in any medium, provided that you conspicuously and
199 | appropriately publish on each copy an appropriate copyright notice;
200 | keep intact all notices stating that this License and any
201 | non-permissive terms added in accord with section 7 apply to the code;
202 | keep intact all notices of the absence of any warranty; and give all
203 | recipients a copy of this License along with the Program.
204 |
205 | You may charge any price or no price for each copy that you convey,
206 | and you may offer support or warranty protection for a fee.
207 |
208 | 5. Conveying Modified Source Versions.
209 |
210 | You may convey a work based on the Program, or the modifications to
211 | produce it from the Program, in the form of source code under the
212 | terms of section 4, provided that you also meet all of these conditions:
213 |
214 | a) The work must carry prominent notices stating that you modified
215 | it, and giving a relevant date.
216 |
217 | b) The work must carry prominent notices stating that it is
218 | released under this License and any conditions added under section
219 | 7. This requirement modifies the requirement in section 4 to
220 | "keep intact all notices".
221 |
222 | c) You must license the entire work, as a whole, under this
223 | License to anyone who comes into possession of a copy. This
224 | License will therefore apply, along with any applicable section 7
225 | additional terms, to the whole of the work, and all its parts,
226 | regardless of how they are packaged. This License gives no
227 | permission to license the work in any other way, but it does not
228 | invalidate such permission if you have separately received it.
229 |
230 | d) If the work has interactive user interfaces, each must display
231 | Appropriate Legal Notices; however, if the Program has interactive
232 | interfaces that do not display Appropriate Legal Notices, your
233 | work need not make them do so.
234 |
235 | A compilation of a covered work with other separate and independent
236 | works, which are not by their nature extensions of the covered work,
237 | and which are not combined with it such as to form a larger program,
238 | in or on a volume of a storage or distribution medium, is called an
239 | "aggregate" if the compilation and its resulting copyright are not
240 | used to limit the access or legal rights of the compilation's users
241 | beyond what the individual works permit. Inclusion of a covered work
242 | in an aggregate does not cause this License to apply to the other
243 | parts of the aggregate.
244 |
245 | 6. Conveying Non-Source Forms.
246 |
247 | You may convey a covered work in object code form under the terms
248 | of sections 4 and 5, provided that you also convey the
249 | machine-readable Corresponding Source under the terms of this License,
250 | in one of these ways:
251 |
252 | a) Convey the object code in, or embodied in, a physical product
253 | (including a physical distribution medium), accompanied by the
254 | Corresponding Source fixed on a durable physical medium
255 | customarily used for software interchange.
256 |
257 | b) Convey the object code in, or embodied in, a physical product
258 | (including a physical distribution medium), accompanied by a
259 | written offer, valid for at least three years and valid for as
260 | long as you offer spare parts or customer support for that product
261 | model, to give anyone who possesses the object code either (1) a
262 | copy of the Corresponding Source for all the software in the
263 | product that is covered by this License, on a durable physical
264 | medium customarily used for software interchange, for a price no
265 | more than your reasonable cost of physically performing this
266 | conveying of source, or (2) access to copy the
267 | Corresponding Source from a network server at no charge.
268 |
269 | c) Convey individual copies of the object code with a copy of the
270 | written offer to provide the Corresponding Source. This
271 | alternative is allowed only occasionally and noncommercially, and
272 | only if you received the object code with such an offer, in accord
273 | with subsection 6b.
274 |
275 | d) Convey the object code by offering access from a designated
276 | place (gratis or for a charge), and offer equivalent access to the
277 | Corresponding Source in the same way through the same place at no
278 | further charge. You need not require recipients to copy the
279 | Corresponding Source along with the object code. If the place to
280 | copy the object code is a network server, the Corresponding Source
281 | may be on a different server (operated by you or a third party)
282 | that supports equivalent copying facilities, provided you maintain
283 | clear directions next to the object code saying where to find the
284 | Corresponding Source. Regardless of what server hosts the
285 | Corresponding Source, you remain obligated to ensure that it is
286 | available for as long as needed to satisfy these requirements.
287 |
288 | e) Convey the object code using peer-to-peer transmission, provided
289 | you inform other peers where the object code and Corresponding
290 | Source of the work are being offered to the general public at no
291 | charge under subsection 6d.
292 |
293 | A separable portion of the object code, whose source code is excluded
294 | from the Corresponding Source as a System Library, need not be
295 | included in conveying the object code work.
296 |
297 | A "User Product" is either (1) a "consumer product", which means any
298 | tangible personal property which is normally used for personal, family,
299 | or household purposes, or (2) anything designed or sold for incorporation
300 | into a dwelling. In determining whether a product is a consumer product,
301 | doubtful cases shall be resolved in favor of coverage. For a particular
302 | product received by a particular user, "normally used" refers to a
303 | typical or common use of that class of product, regardless of the status
304 | of the particular user or of the way in which the particular user
305 | actually uses, or expects or is expected to use, the product. A product
306 | is a consumer product regardless of whether the product has substantial
307 | commercial, industrial or non-consumer uses, unless such uses represent
308 | the only significant mode of use of the product.
309 |
310 | "Installation Information" for a User Product means any methods,
311 | procedures, authorization keys, or other information required to install
312 | and execute modified versions of a covered work in that User Product from
313 | a modified version of its Corresponding Source. The information must
314 | suffice to ensure that the continued functioning of the modified object
315 | code is in no case prevented or interfered with solely because
316 | modification has been made.
317 |
318 | If you convey an object code work under this section in, or with, or
319 | specifically for use in, a User Product, and the conveying occurs as
320 | part of a transaction in which the right of possession and use of the
321 | User Product is transferred to the recipient in perpetuity or for a
322 | fixed term (regardless of how the transaction is characterized), the
323 | Corresponding Source conveyed under this section must be accompanied
324 | by the Installation Information. But this requirement does not apply
325 | if neither you nor any third party retains the ability to install
326 | modified object code on the User Product (for example, the work has
327 | been installed in ROM).
328 |
329 | The requirement to provide Installation Information does not include a
330 | requirement to continue to provide support service, warranty, or updates
331 | for a work that has been modified or installed by the recipient, or for
332 | the User Product in which it has been modified or installed. Access to a
333 | network may be denied when the modification itself materially and
334 | adversely affects the operation of the network or violates the rules and
335 | protocols for communication across the network.
336 |
337 | Corresponding Source conveyed, and Installation Information provided,
338 | in accord with this section must be in a format that is publicly
339 | documented (and with an implementation available to the public in
340 | source code form), and must require no special password or key for
341 | unpacking, reading or copying.
342 |
343 | 7. Additional Terms.
344 |
345 | "Additional permissions" are terms that supplement the terms of this
346 | License by making exceptions from one or more of its conditions.
347 | Additional permissions that are applicable to the entire Program shall
348 | be treated as though they were included in this License, to the extent
349 | that they are valid under applicable law. If additional permissions
350 | apply only to part of the Program, that part may be used separately
351 | under those permissions, but the entire Program remains governed by
352 | this License without regard to the additional permissions.
353 |
354 | When you convey a copy of a covered work, you may at your option
355 | remove any additional permissions from that copy, or from any part of
356 | it. (Additional permissions may be written to require their own
357 | removal in certain cases when you modify the work.) You may place
358 | additional permissions on material, added by you to a covered work,
359 | for which you have or can give appropriate copyright permission.
360 |
361 | Notwithstanding any other provision of this License, for material you
362 | add to a covered work, you may (if authorized by the copyright holders of
363 | that material) supplement the terms of this License with terms:
364 |
365 | a) Disclaiming warranty or limiting liability differently from the
366 | terms of sections 15 and 16 of this License; or
367 |
368 | b) Requiring preservation of specified reasonable legal notices or
369 | author attributions in that material or in the Appropriate Legal
370 | Notices displayed by works containing it; or
371 |
372 | c) Prohibiting misrepresentation of the origin of that material, or
373 | requiring that modified versions of such material be marked in
374 | reasonable ways as different from the original version; or
375 |
376 | d) Limiting the use for publicity purposes of names of licensors or
377 | authors of the material; or
378 |
379 | e) Declining to grant rights under trademark law for use of some
380 | trade names, trademarks, or service marks; or
381 |
382 | f) Requiring indemnification of licensors and authors of that
383 | material by anyone who conveys the material (or modified versions of
384 | it) with contractual assumptions of liability to the recipient, for
385 | any liability that these contractual assumptions directly impose on
386 | those licensors and authors.
387 |
388 | All other non-permissive additional terms are considered "further
389 | restrictions" within the meaning of section 10. If the Program as you
390 | received it, or any part of it, contains a notice stating that it is
391 | governed by this License along with a term that is a further
392 | restriction, you may remove that term. If a license document contains
393 | a further restriction but permits relicensing or conveying under this
394 | License, you may add to a covered work material governed by the terms
395 | of that license document, provided that the further restriction does
396 | not survive such relicensing or conveying.
397 |
398 | If you add terms to a covered work in accord with this section, you
399 | must place, in the relevant source files, a statement of the
400 | additional terms that apply to those files, or a notice indicating
401 | where to find the applicable terms.
402 |
403 | Additional terms, permissive or non-permissive, may be stated in the
404 | form of a separately written license, or stated as exceptions;
405 | the above requirements apply either way.
406 |
407 | 8. Termination.
408 |
409 | You may not propagate or modify a covered work except as expressly
410 | provided under this License. Any attempt otherwise to propagate or
411 | modify it is void, and will automatically terminate your rights under
412 | this License (including any patent licenses granted under the third
413 | paragraph of section 11).
414 |
415 | However, if you cease all violation of this License, then your
416 | license from a particular copyright holder is reinstated (a)
417 | provisionally, unless and until the copyright holder explicitly and
418 | finally terminates your license, and (b) permanently, if the copyright
419 | holder fails to notify you of the violation by some reasonable means
420 | prior to 60 days after the cessation.
421 |
422 | Moreover, your license from a particular copyright holder is
423 | reinstated permanently if the copyright holder notifies you of the
424 | violation by some reasonable means, this is the first time you have
425 | received notice of violation of this License (for any work) from that
426 | copyright holder, and you cure the violation prior to 30 days after
427 | your receipt of the notice.
428 |
429 | Termination of your rights under this section does not terminate the
430 | licenses of parties who have received copies or rights from you under
431 | this License. If your rights have been terminated and not permanently
432 | reinstated, you do not qualify to receive new licenses for the same
433 | material under section 10.
434 |
435 | 9. Acceptance Not Required for Having Copies.
436 |
437 | You are not required to accept this License in order to receive or
438 | run a copy of the Program. Ancillary propagation of a covered work
439 | occurring solely as a consequence of using peer-to-peer transmission
440 | to receive a copy likewise does not require acceptance. However,
441 | nothing other than this License grants you permission to propagate or
442 | modify any covered work. These actions infringe copyright if you do
443 | not accept this License. Therefore, by modifying or propagating a
444 | covered work, you indicate your acceptance of this License to do so.
445 |
446 | 10. Automatic Licensing of Downstream Recipients.
447 |
448 | Each time you convey a covered work, the recipient automatically
449 | receives a license from the original licensors, to run, modify and
450 | propagate that work, subject to this License. You are not responsible
451 | for enforcing compliance by third parties with this License.
452 |
453 | An "entity transaction" is a transaction transferring control of an
454 | organization, or substantially all assets of one, or subdividing an
455 | organization, or merging organizations. If propagation of a covered
456 | work results from an entity transaction, each party to that
457 | transaction who receives a copy of the work also receives whatever
458 | licenses to the work the party's predecessor in interest had or could
459 | give under the previous paragraph, plus a right to possession of the
460 | Corresponding Source of the work from the predecessor in interest, if
461 | the predecessor has it or can get it with reasonable efforts.
462 |
463 | You may not impose any further restrictions on the exercise of the
464 | rights granted or affirmed under this License. For example, you may
465 | not impose a license fee, royalty, or other charge for exercise of
466 | rights granted under this License, and you may not initiate litigation
467 | (including a cross-claim or counterclaim in a lawsuit) alleging that
468 | any patent claim is infringed by making, using, selling, offering for
469 | sale, or importing the Program or any portion of it.
470 |
471 | 11. Patents.
472 |
473 | A "contributor" is a copyright holder who authorizes use under this
474 | License of the Program or a work on which the Program is based. The
475 | work thus licensed is called the contributor's "contributor version".
476 |
477 | A contributor's "essential patent claims" are all patent claims
478 | owned or controlled by the contributor, whether already acquired or
479 | hereafter acquired, that would be infringed by some manner, permitted
480 | by this License, of making, using, or selling its contributor version,
481 | but do not include claims that would be infringed only as a
482 | consequence of further modification of the contributor version. For
483 | purposes of this definition, "control" includes the right to grant
484 | patent sublicenses in a manner consistent with the requirements of
485 | this License.
486 |
487 | Each contributor grants you a non-exclusive, worldwide, royalty-free
488 | patent license under the contributor's essential patent claims, to
489 | make, use, sell, offer for sale, import and otherwise run, modify and
490 | propagate the contents of its contributor version.
491 |
492 | In the following three paragraphs, a "patent license" is any express
493 | agreement or commitment, however denominated, not to enforce a patent
494 | (such as an express permission to practice a patent or covenant not to
495 | sue for patent infringement). To "grant" such a patent license to a
496 | party means to make such an agreement or commitment not to enforce a
497 | patent against the party.
498 |
499 | If you convey a covered work, knowingly relying on a patent license,
500 | and the Corresponding Source of the work is not available for anyone
501 | to copy, free of charge and under the terms of this License, through a
502 | publicly available network server or other readily accessible means,
503 | then you must either (1) cause the Corresponding Source to be so
504 | available, or (2) arrange to deprive yourself of the benefit of the
505 | patent license for this particular work, or (3) arrange, in a manner
506 | consistent with the requirements of this License, to extend the patent
507 | license to downstream recipients. "Knowingly relying" means you have
508 | actual knowledge that, but for the patent license, your conveying the
509 | covered work in a country, or your recipient's use of the covered work
510 | in a country, would infringe one or more identifiable patents in that
511 | country that you have reason to believe are valid.
512 |
513 | If, pursuant to or in connection with a single transaction or
514 | arrangement, you convey, or propagate by procuring conveyance of, a
515 | covered work, and grant a patent license to some of the parties
516 | receiving the covered work authorizing them to use, propagate, modify
517 | or convey a specific copy of the covered work, then the patent license
518 | you grant is automatically extended to all recipients of the covered
519 | work and works based on it.
520 |
521 | A patent license is "discriminatory" if it does not include within
522 | the scope of its coverage, prohibits the exercise of, or is
523 | conditioned on the non-exercise of one or more of the rights that are
524 | specifically granted under this License. You may not convey a covered
525 | work if you are a party to an arrangement with a third party that is
526 | in the business of distributing software, under which you make payment
527 | to the third party based on the extent of your activity of conveying
528 | the work, and under which the third party grants, to any of the
529 | parties who would receive the covered work from you, a discriminatory
530 | patent license (a) in connection with copies of the covered work
531 | conveyed by you (or copies made from those copies), or (b) primarily
532 | for and in connection with specific products or compilations that
533 | contain the covered work, unless you entered into that arrangement,
534 | or that patent license was granted, prior to 28 March 2007.
535 |
536 | Nothing in this License shall be construed as excluding or limiting
537 | any implied license or other defenses to infringement that may
538 | otherwise be available to you under applicable patent law.
539 |
540 | 12. No Surrender of Others' Freedom.
541 |
542 | If conditions are imposed on you (whether by court order, agreement or
543 | otherwise) that contradict the conditions of this License, they do not
544 | excuse you from the conditions of this License. If you cannot convey a
545 | covered work so as to satisfy simultaneously your obligations under this
546 | License and any other pertinent obligations, then as a consequence you may
547 | not convey it at all. For example, if you agree to terms that obligate you
548 | to collect a royalty for further conveying from those to whom you convey
549 | the Program, the only way you could satisfy both those terms and this
550 | License would be to refrain entirely from conveying the Program.
551 |
552 | 13. Use with the GNU Affero General Public License.
553 |
554 | Notwithstanding any other provision of this License, you have
555 | permission to link or combine any covered work with a work licensed
556 | under version 3 of the GNU Affero General Public License into a single
557 | combined work, and to convey the resulting work. The terms of this
558 | License will continue to apply to the part which is the covered work,
559 | but the special requirements of the GNU Affero General Public License,
560 | section 13, concerning interaction through a network will apply to the
561 | combination as such.
562 |
563 | 14. Revised Versions of this License.
564 |
565 | The Free Software Foundation may publish revised and/or new versions of
566 | the GNU General Public License from time to time. Such new versions will
567 | be similar in spirit to the present version, but may differ in detail to
568 | address new problems or concerns.
569 |
570 | Each version is given a distinguishing version number. If the
571 | Program specifies that a certain numbered version of the GNU General
572 | Public License "or any later version" applies to it, you have the
573 | option of following the terms and conditions either of that numbered
574 | version or of any later version published by the Free Software
575 | Foundation. If the Program does not specify a version number of the
576 | GNU General Public License, you may choose any version ever published
577 | by the Free Software Foundation.
578 |
579 | If the Program specifies that a proxy can decide which future
580 | versions of the GNU General Public License can be used, that proxy's
581 | public statement of acceptance of a version permanently authorizes you
582 | to choose that version for the Program.
583 |
584 | Later license versions may give you additional or different
585 | permissions. However, no additional obligations are imposed on any
586 | author or copyright holder as a result of your choosing to follow a
587 | later version.
588 |
589 | 15. Disclaimer of Warranty.
590 |
591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
599 |
600 | 16. Limitation of Liability.
601 |
602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
610 | SUCH DAMAGES.
611 |
612 | 17. Interpretation of Sections 15 and 16.
613 |
614 | If the disclaimer of warranty and limitation of liability provided
615 | above cannot be given local legal effect according to their terms,
616 | reviewing courts shall apply local law that most closely approximates
617 | an absolute waiver of all civil liability in connection with the
618 | Program, unless a warranty or assumption of liability accompanies a
619 | copy of the Program in return for a fee.
620 |
621 | END OF TERMS AND CONDITIONS
622 |
623 | How to Apply These Terms to Your New Programs
624 |
625 | If you develop a new program, and you want it to be of the greatest
626 | possible use to the public, the best way to achieve this is to make it
627 | free software which everyone can redistribute and change under these terms.
628 |
629 | To do so, attach the following notices to the program. It is safest
630 | to attach them to the start of each source file to most effectively
631 | state the exclusion of warranty; and each file should have at least
632 | the "copyright" line and a pointer to where the full notice is found.
633 |
634 |
635 | Copyright (C)
636 |
637 | This program is free software: you can redistribute it and/or modify
638 | it under the terms of the GNU General Public License as published by
639 | the Free Software Foundation, either version 3 of the License, or
640 | (at your option) any later version.
641 |
642 | This program is distributed in the hope that it will be useful,
643 | but WITHOUT ANY WARRANTY; without even the implied warranty of
644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
645 | GNU General Public License for more details.
646 |
647 | You should have received a copy of the GNU General Public License
648 | along with this program. If not, see .
649 |
650 | Also add information on how to contact you by electronic and paper mail.
651 |
652 | If the program does terminal interaction, make it output a short
653 | notice like this when it starts in an interactive mode:
654 |
655 | Copyright (C)
656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
657 | This is free software, and you are welcome to redistribute it
658 | under certain conditions; type `show c' for details.
659 |
660 | The hypothetical commands `show w' and `show c' should show the appropriate
661 | parts of the General Public License. Of course, your program's commands
662 | might be different; for a GUI interface, you would use an "about box".
663 |
664 | You should also get your employer (if you work as a programmer) or school,
665 | if any, to sign a "copyright disclaimer" for the program, if necessary.
666 | For more information on this, and how to apply and follow the GNU GPL, see
667 | .
668 |
669 | The GNU General Public License does not permit incorporating your program
670 | into proprietary programs. If your program is a subroutine library, you
671 | may consider it more useful to permit linking proprietary applications with
672 | the library. If this is what you want to do, use the GNU Lesser General
673 | Public License instead of this License. But first, please read
674 | .
675 |
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # Bayesian Model Updating Tutorials:
2 |
3 | Bayesian Model Updating is a technique which casts the model updating problem in the form of a Bayesian Inference. There have been 3 popular advanced Monte Carlo sampling techniques which are adopted by researchers to address Bayesian Model Updating problems and make the necessary estimations of the epistemic parameter(s). These 3 techniques are:
4 |
5 | * Markov Chain Monte Carlo [(MCMC)](https://doi.org/10.1093/biomet/57.1.97)
6 | * Transitional Markov Chain Monte Carlo [(TMCMC)](https://doi.org/10.1061/(ASCE)0733-9399(2007)133:7(816))
7 | * Sequential Monte Carlo [(SMC)](https://www.jstor.org/stable/3879283)
8 |
9 | In this repository, 3 tutorials are presented to enable users to understand how the advanced Monte Carlo techniques are implemented in addressing various Bayesian Model Updating problems. The following tutorials are (in order of increasing difficulty):
10 |
11 | * 1-Dimensional Linear Static System
12 | * 1-Dimensional Simple Harmonic Oscillator
13 | * 2-Dimensional Eigenvalue Problem
14 |
15 | ## Tutorials:
16 |
17 | ### 1) 1-Dimensional Linear Static System:
18 |
19 | This tutorial presents a simple static Spring-Mass system. In this set-up, the spring is assumed to obey [Hooke's Law](http://latex.codecogs.com/svg.latex?F%3D-k%5Ccdot%7Bd%7D) model whereby the restoring force of the spring, F, is linearly proportional to the length of its displacement from rest length, d. The elasticity constant of the spring is k. This study, seeks to realize two objectives:
20 |
21 | 1. To compare the estimation the epistemic parameter k;
22 |
23 | 2. To compare the model updating results obtained through the use of MCMC, TMCMC, and SMC.
24 |
25 | ### 2) 1-Dimensional Simple Harmonic Oscillator:
26 |
27 | This tutorial presents a simple harmonic oscillator system. In this set-up, the natural oscillating frequency of the ocillator, F, obeys the [Simple Harmonic Frequency](http://latex.codecogs.com/svg.latex?F%3D%5Csqrt%7B%5Cfrac%7Bk%7D%7Bm%7D%7D) model whereby F is defined as the square-root of the ratio between the elasticity constant of the spring, k, and the mass of the body attached to the oscillator, m. This study, seeks to realize two objectives:
28 |
29 | 1. To compare the estimation the epistemic parameter k;
30 |
31 | 2. To compare the model updating results obtained through the use of MCMC, TMCMC, and SMC.
32 |
33 | ### 3) 2-Dimensional Eigenvalue Problem:
34 |
35 | This tutorial presents a 2-by-2 square [matrix](http://latex.codecogs.com/svg.latex?%5Cbegin%7Bpmatrix%7D%0D%0A%7B%5Ctheta_1%7D%2B%7B%5Ctheta_2%7D%26-%7B%5Ctheta_2%7D%5C%5C-%7B%5Ctheta_2%7D%26%7B%5Ctheta_2%7D%5C%5C%0D%0A%5Cend%7Bpmatrix%7D) in which there exists two distinct real eignvalue solutions. The matrix elements here are defined by two epistemic parameters: Theta 1 and Theta 2. This tutorial seeks to achieve three objectives:
36 |
37 | 1. To observe the performance of each of the advanced Monte Carlo samplers in obtaining samples from a 2-dimensional, bi-modal posterior distribution;
38 |
39 | 2. To estimate the solutions to the epistemic parameters: Theta 1 and Theta 2;
40 |
41 | 3. To compare the model updating results obtained through the use of MCMC, TMCMC, and SMC.
42 |
43 | ### 4) Alternative TMCMC Transition criteria:
44 |
45 | The work explores a possible alternative transitional criteria for the TMCMC sampler involving the use of the Effective sample size metric. The alternative transitional criteria is such that in transiting from one transitional distribution to another, the Effective sample size has to be half the total sample size. The alternative TMCMC sampler is referred to as the TMCMC-II sampler and it will be implemented on a SDoF structure subjected to an unknown Coulomb friction. The code to be executed is named: "example_SDOF_System_Coulomb_Friction_numerical.m"
46 |
47 | The work was presented at the 33rd European Safety and Reliability Conference (ESREL 2023) held in Southampton, United Kingdom.
48 |
49 | ## Reference(s):
50 | * A. Lye, A. Cicirello, and E. Patelli (2021). Sampling methods for solving Bayesian model updating problems: A tutorial. *Mechanical Systems and Signal Processing, 159*, 107760. doi: [10.1016/j.ymssp.2021.107760](https://doi.org/10.1016/j.ymssp.2021.107760)
51 | * A. Lye, and L. Marino (2023). An investigation into an alternative transition criterion of the Transitional Markov Chain Monte Carlo method for Bayesian model updating. *In Proceedings of the 33rd European Safety and Reliability Conference, 1*. doi: [10.3850/978-981-18-8071-1_P331-cd](https://doi.org/10.3850/978-981-18-8071-1_P331-cd)
52 |
53 | ## Author:
54 | * Name: Adolphus Lye
55 | * Contact: adolphus.lye@liverpool.ac.uk
56 | * Affiliation: Insitute for Risk and Uncertainty, University of Liverpool
57 |
--------------------------------------------------------------------------------