sot::Optimizer Class Reference

The surrogate optimization algorithm. More...

#include <optimizer.h>

Public Member Functions
	Optimizer (std::shared_ptr< Problem > &data, std::shared_ptr< ExpDesign > &expDes, std::shared_ptr< Surrogate > &surf, std::shared_ptr< Sampling > &sampling, int maxEvals)
	Constructor. More...
	Optimizer (std::shared_ptr< Problem > &data, std::shared_ptr< ExpDesign > &expDes, std::shared_ptr< Surrogate > &surf, std::shared_ptr< Sampling > &sampling, int maxEvals, int numThreads)
	Constructor. More...
Result	run ()
	Runs the optimization algorithm. More...

Protected Member Functions
void	evalBatch (const mat &batch, vec &funVals)
	Evalaute a batch of points in parallel. More...

Protected Attributes
std::shared_ptr< Problem >	mData
std::shared_ptr< ExpDesign >	mExpDes
std::shared_ptr< Surrogate >	mSurf
std::shared_ptr< Sampling >	mSampling
const double	mSigmaMax = 0.2
const double	mSigmaMin = 0.005
int	mSuccTol = 3
int	mFailTol
int	mMaxEvals
int	mNumEvals
int	mInitPoints
int	mDim
vec	mxLow
vec	mxUp
std::string	mName = "Surrogate Optimizer"
int	mNumThreads
int	mEvalCount = 0
std::mutex	mMutex

Detailed Description

The surrogate optimization algorithm.

This is a general surrogate optimization algorithm that uses the supplied experimental design, surrogate model, and adaptive sampling to minimize the given optimization problem. The optimization algorithm starts by searching far from previously generated points and start searching closer and closer to the best solution found so far when not enough improvement is made. The algorithm restarts with a new experimental design when the search radius gets too small.

Author

David Eriksson, dme65.nosp@m.@cor.nosp@m.nell..nosp@m.edu

Constructor & Destructor Documentation

sot::Optimizer::Optimizer ( std::shared_ptr< Problem > &  data, std::shared_ptr< ExpDesign > &  expDes, std::shared_ptr< Surrogate > &  surf, std::shared_ptr< Sampling > &  sampling, int  maxEvals  )

inline

Constructor.

Parameters

data	A shared pointer to the optimization problem
expDes	A shared pointer to the experimental design
surf	A shared pointer to the surrogate model
sampling	A shared pointer to the adaptive sampling
maxEvals	Evaluation budget

Exceptions

std::logic_error

If size of experimental design exceeds the evaluation budget

sot::Optimizer::Optimizer ( std::shared_ptr< Problem > &  data, std::shared_ptr< ExpDesign > &  expDes, std::shared_ptr< Surrogate > &  surf, std::shared_ptr< Sampling > &  sampling, int  maxEvals, int  numThreads  )

inline

Constructor.

Parameters

data	A shared pointer to the optimization problem
expDes	A shared pointer to the experimental design
surf	A shared pointer to the surrogate model
sampling	A shared pointer to the adaptive sampling
maxEvals	Evaluation budget
numThreads	Number of threads to use for parallel evaluations

Exceptions

std::logic_error

If size of experimental design exceeds the evaluation budget

Member Function Documentation

void sot::Optimizer::evalBatch ( const mat &  batch, vec &  funVals  )

inlineprotected

Evalaute a batch of points in parallel.

Parameters

batch	Batch of points to be evaluated
funVals	Vector to write the function values to

Result sot::Optimizer::run ( )

inline

Runs the optimization algorithm.

Returns

A Result object with the results from the run

Member Data Documentation

std::shared_ptr<Problem> sot::Optimizer::mData

protected

A shared pointer to the optimization problem

int sot::Optimizer::mDim

protected

Number of dimensions (extracted from mData)

int sot::Optimizer::mEvalCount = 0

protected

Evaluation counter for evalauting batches

std::shared_ptr<ExpDesign> sot::Optimizer::mExpDes

protected

A shared pointer to the experimental design

int sot::Optimizer::mFailTol

protected

After mSuccTol consecutive iteration without improvement the search radius is decreased

int sot::Optimizer::mInitPoints

protected

Points in the initial design (extracted from mExpDes)

int sot::Optimizer::mMaxEvals

protected

Evaluation budget

std::mutex sot::Optimizer::mMutex

protected

Mutex for assigning evaluations to the threads

std::string sot::Optimizer::mName = "Surrogate Optimizer"

protected

Strategy name

int sot::Optimizer::mNumEvals

protected

Evaluations carried out

int sot::Optimizer::mNumThreads

protected

Number of threads

std::shared_ptr<Sampling> sot::Optimizer::mSampling

protected

A shared pointer to the adpative sampling

const double sot::Optimizer::mSigmaMax = 0.2

protected

Largest search radius w.r.t. the unix box

const double sot::Optimizer::mSigmaMin = 0.005

protected

Smallest search radius w.r.t. the unix box

int sot::Optimizer::mSuccTol = 3

protected

After mSuccTol consecutive improvements the search radius is increased

std::shared_ptr<Surrogate> sot::Optimizer::mSurf

protected

A shared pointer to the surrogate model

vec sot::Optimizer::mxLow

protected

Lower variable bounds (extracted from mData)

vec sot::Optimizer::mxUp

protected

Upper variable bounds (extracted from mData)

---

The documentation for this class was generated from the following file:

• include/sot_bits/optimizer.h