Train Map

template<typename MemorySpace>
double mpart::TrainMap(std::shared_ptr<ConditionalMapBase<MemorySpace>> map, std::shared_ptr<MapObjective<MemorySpace>> objective, TrainOptions options)

Function to train a map inplace given an objective and optimization options.

Parameters:

• map – Map to optimize (inplace)

• objective – MapObjective to optimize over

• options – Options for optimizing the map

struct TrainOptions

TrainOptions adds options for training your map, with fields largely based on NLopt settings. For documentation of such fields, see NLOpt.

Subclassed by mpart::ATMOptions

Public Functions

inline std::string String()

Create a string representation of these training options (helpful for bindings)

Returns:

std::string Every option value in this struct

Public Members

std::string opt_alg = "LD_SLSQP"

NLOpt: Optimization Algorithm to use

double opt_stopval = -std::numeric_limits<double>::infinity()

NLOpt: Lower bound on optimizer

double opt_ftol_rel = 1e-3

NLOpt: Relative tolerance on function value change

double opt_ftol_abs = 1e-3

NLOpt: Absolute tolerance of function value change

double opt_xtol_rel = 1e-4

NLOpt: Relative tolerance of minimizer value change

double opt_xtol_abs = 1e-4

NLOpt: Absolute tolerance of minimizer value change

int opt_maxeval = 1000

NLOpt: Maximum number of evaluations of function to optimize

double opt_maxtime = std::numeric_limits<double>::infinity()

NLOpt: Maximum amount of time to spend optimizing

int verbose = 0

Verbosity of map training (1: verbose, 2: debug)