Parameter estimation with more data sets
7 views (last 30 days)
Is it possible to do a parameter estimation using 2 or 3 data sets? I have 3 models which differ in one parameter, say T. Right now, I do the fits for each data set and the corresponding model and then compare the results. Can I do a parameter estimation which considers all three data sets and the 3 variants, and outputs only one set of optimized parameters?
Arthur Goldsipe on 5 May 2011
Here's additional information on how you can fit simulations to the same data set using lsqnonlin. You can also find several lsqnonlin/lsqcurvefit examples in the Optimization Toolbox documentation: http://www.mathworks.com/help/toolbox/optim/ug/brn4noo.html#brp3l6v-1
You will also probably want to read more about using the SimBiology command-line. In particular, I recommend reading about variants, as that is a convenient way to modify parameter values for various simulations. If performance is an issue, then also read about the sbioaccelerate command.
Here's an outline of what your regression function might look for lsqnonlin.
function residuals = simulate(pEstimates, model, pVariant, tVariants)
% pEstimates is a vector of of the current parameter estimates
% model is a SimBiology model
% pVariant is a variant for setting the parameters on the model
% tVariant is a vector of variants, one for each value of T
% Update pVariant with the current estimates (in pEstimates)
% For each tVariant, simulate the model using this variant AND pVariant
% Resample the simulation data to the observed times
% Calculate the residual for each observation
More Answers (2)
Jarrod Rivituso on 26 Apr 2011
Are you referring to parameter estimation tasks with Simulink models?
If so, I believe it is. You just add each data set under "Transient Data", and then when you start a new Estimation, you can select all of the data sets in the "Data Sets" tab.
Arthur Goldsipe on 27 Apr 2011
sbioparamestim only allows you to estimate one set of parameters using one set of data. I think the easiest way to simultaneously fit all the data sets would be to write your own regression function for use with a general fitting function (such as lsqcurvefit). Are you also trying to estimate the parameter T, or is that known in advance? Either way, you should be able to write an appropriate regression function.