Determines how many training samples are taken. Value MUST be set taking into account the full system's dimension or at least the bounding box of the samples!
Not setting this value in implementing subclasses causes KerMor's ODE solvers to (possibly) produce wrong results due to wrong assumptions on the time dependence of the core function.
Not setting this value in implementing subclasses causes KerMor's ODE solvers to (possibly) produce wrong results due to wrong assumptions on the time dependence of the core function.
Some output conversion matrices are time dependent. This property must be set to the correct value in order for the output conversion to work correctly.
Using more \(\mu\) parameter vector elements in the kernel approximations than actually required in the core function \(f\) introduces an additional dependency of the nonlinearity on extra parameters which is not given in the full model's core function.
If manual sampling is used, this property builds the basic set of parameter samples to be used for offline computations and hence must be chosen with maximum care.
Too large steps due to high time-step distances in the passed times vector \(t\) may lead to errorneous results. This property limits the maximum time-step size used in the implementations. Set to [] in order to rely on the times \(t\).