Syntax
>>> GRID SEARCH (UNSORTED) (: ninval1 (ninval2 (ninval3)) / FILE: filename)
or
>>> OBJECTIVE (UNSORTED) (: ninval1 (ninval2 (ninval3)) / FILE: filename)
Parent Command
>> OPTION
Subcommand
-
Description
This command evaluates the objective function for a number of specific parameter sets. A arbitrarily long list of parameter sets can be provided on file filename, each row consisting of one parameter set, whereby inactive and tied parameters are to be omitted.
Alternatively, parameter sets are generated internally on a regular grid, mapping out the entire parameter space. In this case, lower and upper bounds must be defined for each parameter using command >>>> RANGE in block > PARAMETER. This range is then subdivided into ninteri intervals by inserting ninvali+1 equally spaced points, generating parameter sets on a regular grid in the n-dimensional parameter space. The objective function is evaluated at each grid point. Evaluating the objective function throughout the entire parameter space is referred to as grid searching. The parameter set, the value of the objective function, and the contribution of each observation type to the objective function is printed to the iTOUGH2 output file. The global minimum is likely to be in the vicinity of the parameter set with the lowest objective function. Furthermore, the information listed in the output file can be used to visually represent and study the topology of the objective function. For example, one-liner can generate a contour plot of the objective function for n=2 which may reveal the presence local minima.
If only one output variable is defined in block > OBSERVATION and no data are provided, this option can also be used in combination with command >>> L1-ESTIMATOR to examine the sensitivity of the output variable over an extensive parameter range.
Keyword UNSORTED can be used in PVM applications to improve efficiency. The output list must then be sorted in a post-processing step.
Example
> COMPUTATION
>> OPTION
>>> GRID SEARCH! subdivide 2D-parameter space into: 20 10 intervals
<<<
<<
See Also
>>> ANNEAL | >>> LEVENBERG-MARQUARDT | >>> GAUSS-NEWTON | >>> TORNADO