Optimization and MCMC can use a new :SYNTHETIC keyword to generate synthetic data and draws from priors.
As of v10.5, this capability is integrated into the optimization control editor interface, or you can specify the keyword by editing the .voc:
You can use the SYNTHETIC keyword with the optimizer OFF to generate synthetics without running an optimization or MCMC.
Syntax
The format is :SYNTHETIC=#, where number is a fraction between -1.0 and +1.0.
If the # argument is 0, the SYNTHETIC option is skipped.
If the # is nonzero, Vensim generates two new outputs:
•Runname_syn_prior.cin containing a random draw of your parameter set from its priors, as they are specified in the parameter list.
oIf no prior is specified, it uses a very broad heavy-tailed distribution, as for the initialization of random multistart runs.
•Runname_syn_data.dat containing synthetic data corresponding with calibration payoff elements
The parameter # controls data generation:
A negative value, -1 < # < 0 causes generation of a synthetic point corresponding with each real point you have supplied in other data; use this to create a synthetic copy of your data. Each time the calibration payoff calculation reaches a model-data comparison where the data is populated (i.e., not NA), a random draw will be computed from the model value, plus a random component using the error distribution specified in the payoff.
A positive value, 0 < # < 1 causes generation of a synthetic point at each SAVEPER; a random draw will be computed from the model value, plus a random component using the error distribution specified in the payoff.
If 0<ABS(#)<1, the data is thinned randomly (useful for creating sparse data). For example,
:SYNTHETIC=0.5
will essentially flip a coin to decide whether to generate a given data point at each SAVEPER.
