%% ALGORITHM SETUP AND INITIALIZATION % This script will be the msot heavily modified section of the code. It % itializes the fit setttings and sets up the data and inital guesses. ALL % SECTIONS THAT NEED TO BE MODIFIED ARE MARKED WITH *** %% DATA INDEXING %Partitions data into the independent variable, dependent %variable and the StDev on the dependent variable. indvar = InsertData_1(:,1); % pulls out indvar data_input_1 = InsertData_1(:,2:2:end); %pulls out data weights_1 = InsertData_1(:,3:2:end); %pulls out weights data_input_2 = InsertData_2(:,1:2:end); %pulls out data weights_2 = InsertData_2(:,2:2:end); %pulls out weights for i=1:size(weights_1,1) %weights should not be equal to zero. This checks weights matrix. for j=1:size(weights_1,2) if weights_1(i,j)<0.01 weights_1(i,j)=0.01; end end end for i=1:size(weights_2,1) %weights should not be equal to zero. This checks weights matrix. for j=1:size(weights_2,2) if weights_2(i,j)<0.01 weights_2(i,j)=0.01; end end end %% PARAMETER GUESSES*** % Defines the intial guesses that the algorithm will use as a starting % point. The GA portion of the algorithm will help reduce how sensitive % the fit is to the inital guessess; however, these guesses need to be % reasonable(perferably within an order of magnitude). Each parameter % is also asigned an upper and lower bound. Note: Upper and Lower % bounds have to be different number % Param 1 % k1 = 1; % Initial guess % k1ub = 500; % Upper bound % k1lb = 0.01; % Lower bound % % k2 = 1; % k2ub = 500; % k2lb = 0.01; % % k3 = 1; % k3ub = 500; % k3lb = 0.01; % % k4 = 1; % k4ub = 500; % k4lb = 0.01; % % % % U_0 = 0.1; % U_0ub = 1; % U_0lb = 0.001; % % I1_0 = 0.1; % I1_0ub = 1; % I1_0lb = 0.001; % % I2_0 = 0.1; % I2_0ub = 1; % I2_0lb = 0.001; % Concatenates params and their bounds into a single matrix *** % keep the format of [x x x x x x...]; % param_vals = [k1 k2 k3 k4 U_0 I1_0 I2_0]; % params_ub = [k1ub k2ub k3ub k4ub U_0ub I1_0ub I2_0ub]; % params_lb = [k1lb k2lb k3lb k4lb U_0lb I1_0lb I2_0lb]; param_vals = [0.153850002869249,0.0966098569539615,2803504949.32173,0.738904110904648,0.157256734041407,0.0641849476554585,10000000000.0000,2.86275545312608,0.159656258881822,0.00100371485116490,0.914666600028426,0.377026909737468,0.116339358500157,0.0430896689185434,0.753692498731245,0.105947141660370,0.147799691437948,0.155686140331968,0.196443559306337,0.607235522494936,0.128960919824300,0.348974939635825,0.471072774195592]; params_ub = [1.00E+10 1.00E+10 1.00E+12 1.00E+10 1.00E+10 1.00E+10 1.00E+12 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10 1.00E+10]; params_lb = [0.001 0.001 1E7 0.001 0.001 0.001 1E7 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001]; % clear k1 k2 k3 k4 U_0 I1_0 I2_0 k1ub k2ub k3ub k4ub U_0ub I1_0ub I2_0ub ... % k1lb k2lb k3lb k4lb U_0lb I1_0lb I2_0lb %% FIT SETTINGS % Iteration Control Vectors:*** %Controls how many cycles the algorithm will perform. Mute unneeded %cycles and create more by copying and editing the index number to %match the cycle number(Cycle #** --> kt_c(**)=1;). Within each cycle %set the number of generations by editing the value of g and control %which parameters will be floated in each cycle(1=float 0=dont float) %Set LSQNL# equal to 1 to turn on, set equal to 0 to turn off %RECOMMENDATION: %The default iteration control vector should consist of a %first cycle that floats all of the fit parameters for 10 generations with %LSQNL turned off, with a second cycle that floats all the parameters for %10 generations with LSQNL turned on. This works for most cases. % Cycle #1 g(1)=10; % g(1) is the first element in the generations matrix which % corresponds to the number of generations to be performed % in the first cycle, 5 indicates how many generations to % perform within the first cycle. % k1_c(1) = 1; % m_c(1) : m is the parameter name, _c indicates the cycles % % matrix, (1) inidicates the first cycle, = 1 indicates this % % is a floating parameter. % k2_c(1) = 1; % k3_c(1) = 1; % k4_c(1) = 1; % U_0_c(1) = 1; % I1_0_c(1) = 1; % I2_0_c(2) = 1; % % NLLS(1)=0; % This defines if the first cycle should perform NLLS (= 1) % or not (= 0). It is generally suggested that NLLS not % be performed on the first cyle. % Cycle #2 g(2)=10; % g(2) is the second element in the generations matrix which % corresponds to the number of generations to be performed in % the second cycle in the the second ... % cycle, 3 indicates how many generations to perform within % the second cycle. % % k1_c(2) = 1; % k2_c(2) = 1; % k3_c(2) = 1; % k4_c(2) = 1; % U_0_c(2) = 1; % I1_0_c(2) = 1; % I2_0_c(2) = 1; % % NLLS(2)=1; % Concatenates iteration control vector*** % format as such [x;x;x;x;x]; (need ; between each term) % CyMat=[k1_c; k2_c; k3_c; k4_c; U_0_c; I1_0_c; I2_0_c; NLLS]'; CyMat = [ones(size(param_vals)) 0 ones(size(param_vals)) 1]; CyMatH=size(CyMat,1); clear k1_c k2_c k3_c k4_c U_0_c I1_0_c I2_0_c NLLS; %% OVERALL SETTINGS: % Stopcrit % Defines the stopping criteria. The algorithm will conclude it has % converged and stop when (1) the mean, absolute coefficient of variance(cv) of the % parameters is less than the stopcrit_params (2) the percent % difference between the chisq at the begining of period and the % best chisq at the end of the period is less than the % stopcrit_chisq (3) the fit has run stopcrit_period periods(this % last stopping criteria is to ensure the fit doesnt get stuck and % go on indefinately) stopcrit_param = 0.01; % cutoff for parameter cv (0.01 is recommended default) stopcrit_chisq = 1; % cutoff for percent difference of chisq(1% is recommended default) stopcrit_period = 10; % cutoff for maximum number of periods (10 is recommend default) % ORnum % Defines the number of optimization routines that each period will run through(3 is the recommended default). ORnum = 3; % Pop_size % Indicates number of individuals within each generation(for 1-5 % parameters 20 is the recommended default, 6-10=200, 11-15=1000, % >15=2000) pop_size = 50000; %% RUN FIGURE SETTINGS: % Run_fig % This controls whether a run figure is generated throughout the % fitting process. This figure shows general fit information, the % current fit of the data, and the change in chisq. Set to 'on' or 'off' run_fig = 'on'; % Axspace % Choose whether the data dsiplayed in the run figure is on a % linear or log scale. Linear 'lin' or Log 'log' axspace = 'lin'; %% GA SETTINGS: % Crossover % Fraction of individuals that are generated from crossover in each % generation(rest of the fraction is generated from mutation)(0.85 is the recommended default) crossover = 0.85; % Stdevs % Defines the normal distribution by which random mutations are % generated(the is fractional, 1=100% of value of the parameter)(1 is the recommended default) stdevs = ones(size(param_vals)); % Best % Number of individuals used to make the best population.(must be % greater than nlls_size) (5 is the recommended default) best = 5; % Mut_probability % This number defines the probability that a parameter value within % an individual will be mutated. When individuals are being % produced by mutation, the probability that the value of parameter % (n) of a specific individual will be produced by mutation is % mut_probability. The corrolary is that the probability of the % value of parameter (n) of a specific individual running through % the mutation process will reach the next generation un-altered is % 1-mut_probability. (0.5 is the recommended default) mut_probability = 0.5; % Elite % This is the number of individuals that advance to the next % generation unaltered. The individuals are chosen by ranking the % Chisq values. The individuals that advance are the top (n) where % the value of "n" is defined by the elite field. The way the % algorithm is written, elite must be less than or equal to best. (3 is the recommended default) elite = 3; % Decay % This controls the rate of change of the standard deviations used % during mutation. The standard deviation changes according to the % reciprocal of the generation number according to % stdev(generationN)=stdev(generation1)*abs((1-decay*(generationN/generationT))) % where generation"N" refers to the generation number and % generation"T" refers to the total number of generations. % Depending on the value of decay the standard deviation changes in % different ways. For decay of zero, standard deviation is % constant. For decay of one, standard deviation shrinks from the % initial standard deviation to zero at a rate of 1/generation. For % a decay between 1 and generations, the standard deviation shrinks % at a rate dictated by decay until decay equals the generation % number then grows. For decay values larger than generations the % standard deviation grows throughout the run. (1 is the recommended default) decay =1; %% NLLS SETTINGS: % AlgSw % Sets which algorithm the lsqnl will utilize. Default setteing is % 0 which uses 'trust-region-reflective'. The other option is 1 for % 'levenberg-marquardt'. For more info type "doc Choosing the % Algorithm" into the command window and look under lsqnonlin algsw = 0; % Nlls_size % Defines the number of individuals that will be refined via NLLS analysis % in each generation where NLLS is being performed.(must be lower % than best) (3 is the recommended default) nlls_size = 1; %% Save Settings save('Start_Workspace.mat')