OEOptimizer1

Attention

This API is currently available in C++ and Python.

class OEOptimizer1

The OEOptimizer1 abstract class defines the interface for optimizing a set of variables for which a function and gradients can be evaluated. Implementations of OEOptimizer1 use both function values and gradients during optimization.

The OEOptimizer1 class defines the following public methods:

• operator()

• GetIterLimit and SetIterLimit

• SetLineMinimize

• GetTolerance and SetTolerance

The following classes derive from this class:

• OEBFGSOpt

• OEConjGradOpt

• OEDFPOpt

• OEFletcherReevesOpt

• OEHestenesStiefelOpt

• OERibierePolakOpt

• OESteepestDescentOpt

operator()

double operator()(OEFunc1 &function, const double *var, double *opt_var)
double operator()(OEFunc1 &function, OECheckpoint1 *check, const double *var,
                  double *opt_var)

These virtual methods define the interface for optimizing a set of variables with the use of evaluated function values. The reference to the function object is passed as a first argument. Initial values of variables are passed in the array var. The optimized variables are returned in the array opt_var. Both operators return the value of the corresponding optimized function value. The second operator takes a pointer to the OECheckpoint1 object that can be used to monitor the progress of an optimization.

GetIterLimit

unsigned GetIterLimit() const

See SetIterLimit.

GetTolerance

double GetTolerance() const

See SetTolerance.

SetIterLimit

void SetIterLimit(unsigned int itmax)

Defines the interface for setting the maximum number of iterations that an optimizer derived from the OEOptimizer0 will attempt while trying to identify a converged minimum function value. Optimization will cease if the iteration limit is reached without finding a converged minimum.

SetLineMinimize

bool SetLineMinimize(const OELineMinimize &linmin)

Allows selection of the line minimization object used by the optimizer.

SetTolerance

void SetTolerance(double t)

Defines the interface for setting the gradient convergence criteria. Depending on the implementation it might be the root mean square gradient, gradient norm or gradient dot product. Optimization will terminate normally if the convergence criteria set by this method is achieved during an optimization.