Optimizer

3) Optimizer

This section selects optimization mode and maps GUI scheme choices to runtime optimizer fields.

3.1 Core optimization controls

These keywords control the optimizer type, GA selection strategy, and convergence thresholds.

Keyword	Default value	Description
optimizer	"ga"	Type of optimizer. Possible values: "ga" or "nelder-mead".
ga_type	"exp"	Genetic algorithm selection strategy. Possible values: "exp" or "tournament".
NMS_start	""	Optional seed generation for NM swarm. For example G0001 is the first generation ran by a GA (which only correspond to models perturbed from the initial model). GT-1 point to the last (pythonic -1) GOAT ensemble, which is the ensemble of top models from all previously ran generation.
n_mdl	500	Number of models per generation.
goat_length	250	Size of top models kept in the GOAT ensemble.
max_gen	10	Maximum generations. Note: use 1 if you only want to perform a "Swarm of Nelder-Mead".

3.2 Nelder-Mead controls

These keywords control Nelder-Mead behavior. They are only used if optimizer is set to "nelder-mead". See SciPy documentation for details on the Nelder-Mead algorithm and its parameters (https://docs.scipy.org/doc/scipy/reference/optimize.minimize-neldermead.html).

Keyword	Default value	Description
nm_fatol	1	NM function absolute tolerance.
nm_xatol	0.5	NM parameter absolute tolerance.
nm_maxiter	0	NM max iterations (0 means solver default behavior where applicable).
nm_maxfev	0	NM max function evaluations (0 means solver default behavior where applicable).
nm_dstep	0.5	Initial simplex scaling step for NM. The simplex is created using a derivative step of every active parameters, plus the initial model.
nm_adaptive	false	Enables adaptive Nelder-Mead variant.

Currently, the derivative step used to build the initial simplex is positive direction only, its size is set by the parameter's class. For multiplicative parameters (if, sfc, mrc, bfc, frequencies) the step is multiplicative and computed in log space: with factor f = uc**nm_dstep, the up step is value * f.

During the final stage of optimization, the Nelder-Mead algorithm is run again with tighter tolerances after a second sensitivity analysis from the previously optimized simplex. These keywords control the final-stage NM behavior. They are only used if optimizer is set to "nelder-mead". See SciPy documentation for details on the Nelder-Mead algorithm and its parameters (https://docs.scipy.org/doc/scipy/reference/optimize.minimize-neldermead.html).

Keyword	Default value	Description
nm_final_fatol	0.05	Final-stage NM tolerance (in defaults, currently not surfaced in GUI controls).
nm_final_xatol	0.005	Final-stage NM parameter tolerance (not surfaced in GUI controls).
nm_final_maxiter	0	Final-stage NM max iterations (not surfaced in GUI controls).
nm_final_maxfev	0	Final-stage NM max evaluations (not surfaced in GUI controls).
nm_final_adaptive	false	Final-stage adaptive flag (not surfaced in GUI controls).