void mainAuto()
{
searchDirection();
eraseDisplay();
searchStrength();
// 3438 ticks = half way point
if (encoderCount >= 3438) {
rotateTwo();
} else {
rotateOne();
}
unRotate();
}
Vector<double> GradientDescent::getMinimalArgument(void)
{
int numberOfVariables = objectiveFunction->getNumberOfVariables();
Vector<double> minimalArgument(numberOfVariables, 0.0);
Vector<double> argument(numberOfVariables, 0.0);
// Evaluation history vector
Vector<double> newEvaluationHistory(maximumNumberOfIterations+1, 0.0);
evaluationHistory = newEvaluationHistory;
// Gradient norm history vector
Vector<double> newGradientNormHistory(maximumNumberOfIterations+1, 0.0);
gradientNormHistory = newGradientNormHistory;
double evaluation = 0.0;
Vector<double> gradient(numberOfVariables, 0.0);
double gradientNorm = 0.0;
Vector<double> searchDirection(numberOfVariables, 0.0);
double initialStepSize = 0.0;
double optimalStepSize = 0.0;
double oldOptimalStepSize = 0.0;
time_t beginningTime, currentTime;
double elapsedTime = 0.0;
// Set beginning time
time(&beginningTime);
std::cout << std::endl
<< "Getting minimal argument with gradient descent..."
<< std::endl;
argument = initialArgument;
// Initial objective function evaluation
evaluation = objectiveFunction->getEvaluation(argument);
evaluationHistory[0] = evaluation;
if(evaluation <= evaluationGoal)
{
std::cout << std::endl
<< "Initial evaluation is less than goal." << std::endl
<< "Initial evaluation: " << evaluation << std::endl;
minimalArgument = argument;
// Print minimal argument to screen
std::cout << std::endl
<< "Minimal argument:" << std::endl;
for(int i = 0; i < numberOfVariables; i++)
{
std::cout << minimalArgument[i] << " ";
}
return(minimalArgument);
}
else
{
std::cout << "Initial evaluation: " << evaluation << std::endl;
}
// Initial objective function gradient
gradient = objectiveFunction->getGradient(argument);
gradientNorm = objectiveFunction->getGradientNorm(gradient);
gradientNormHistory[0] = gradientNorm;
if(gradientNorm <= gradientNormGoal)
{
std::cout << std::endl
<< "Initial gradient norm is less than goal." << std::endl
<< "Initial gradient norm: " << gradientNorm << std::endl;
minimalArgument = argument;
// Print minimal argument to screen
std::cout << std::endl
<< "Minimal argument:" << std::endl;
for(int i = 0; i < numberOfVariables; i++)
{
std::cout << minimalArgument[i] << " ";
}
return(minimalArgument);
}
else
{
std::cout << "Initial gradient norm: " << gradientNorm << std::endl;
}
// Main loop
for(int iteration = 1; iteration <= maximumNumberOfIterations; iteration++)
{
// Get search direction
for(int i = 0; i < numberOfVariables; i++)
{
searchDirection[i] = -1.0*gradient[i];
}
// Get optimal step size
if(iteration == 1)
{
initialStepSize = firstStepSize;
}
else
{
initialStepSize = oldOptimalStepSize;
}
switch(optimalStepSizeMethod)
{
case GoldenSection:
optimalStepSize = getGoldenSectionOptimalStepSize
(initialStepSize, evaluation, argument, searchDirection);
break;
case BrentMethod:
optimalStepSize = getBrentMethodOptimalStepSize
(initialStepSize, evaluation, argument, searchDirection);
break;
}
// Optimal step size stopping criterium
if(optimalStepSize == 0.0)
{
std::cout << std::endl
<< "Iteration " << iteration << ": "
<< "Optimal step size is zero." << std::endl;
std::cout << "Final evaluation: " << evaluation << ";" << std::endl;
std::cout << "Final gradient norm: " << gradientNorm << ";" << std::endl;
break;
}
// Get new free parameters
for (int i = 0; i < numberOfVariables; i++)
{
argument[i] += optimalStepSize*searchDirection[i];
}
// Objective function evaluation
evaluation = objectiveFunction->getEvaluation(argument);
evaluationHistory[iteration] = evaluation;
// Objective function gradient
gradient = objectiveFunction->getGradient(argument);
gradientNorm = objectiveFunction->getGradientNorm(gradient);
gradientNormHistory[iteration] = gradientNorm;
// Stopping Criteria
// Evaluation goal
if (evaluation <= evaluationGoal)
{
std::cout << std::endl
<< "Iteration " << iteration << ": "
<< "Evaluation goal reached." << std::endl;
std::cout << "Evaluation: " << evaluation << ";" << std::endl;
std::cout << "Gradient norm: " << gradientNorm << ";" << std::endl;
break;
}
// Norm of objective function gradient goal
if (gradientNorm <= gradientNormGoal)
{
std::cout << std::endl
<< "Iteration " << iteration << ": "
<< "Gradient norm goal reached."
<< std::endl;
std::cout << "Evaluation: " << evaluation << ";" << std::endl;
std::cout << "Gradient norm: " << gradientNorm << ";" << std::endl;
break;
}
// Maximum optimization time
time(¤tTime);
elapsedTime = difftime(currentTime, beginningTime);
if (elapsedTime >= maximumTime)
{
std::cout << std::endl
<< "Iteration " << iteration << ": "
<< "Maximum optimization time reached."
<< std::endl;
std::cout << "Evaluation: " << evaluation << ";" << std::endl;
std::cout << "Gradient norm: " << gradientNorm << ";" << std::endl;
break;
}
// Maximum number of iterations
if (iteration == maximumNumberOfIterations)
{
std::cout << std::endl
<< "Iteration " << iteration << ": "
<< "Maximum number of iterations reached."
<< std::endl;
std::cout << "Evaluation: " << evaluation << ";" << std::endl;
std::cout << "Gradient norm: " << gradientNorm << ";" << std::endl;
break;
}
// Progress
if(iteration % showPeriod == 0)
{
std::cout << std::endl
<< "Iteration " << iteration << "; " << std::endl;
std::cout << "Evaluation: " << evaluation << ";" << std::endl;
std::cout << "Gradient norm: " << gradientNorm << ";" << std::endl;
}
// Update
oldOptimalStepSize = optimalStepSize;
}
// Set minimal argument
minimalArgument = argument;
// Print minimal argument to screen
std::cout << std::endl
<< "Minimal argument:" << std::endl;
for(int i = 0; i < numberOfVariables; i++)
{
std::cout << minimalArgument[i] << " ";
}
std::cout << std::endl;
return(minimalArgument);
}