patternsearch

Source

Files

Makefile
exploreh.h
exploreline.h
explorelinetest.cpp
explorelinetest.h
explorepd1.h
explorepdtest.cpp
explorepdtest.h
explorequad.h
explorerand.h
exploretest.cpp
exploretest.h
leastsqrs.cpp
leastsqrs.h
lineopgold.h
lineoptimizergold.h
lineoptimizergold2.h
lineoptimizergoldtest.cpp
lineoptimizergoldtest.h
lineoptimizerparabola.h
linepathD1.h
linepath_d1.h
main.cpp
main2.cpp
minexpdim.h
minexpdimtest.cpp
minexpdimtest.h
minimizecomparetest.cpp
minimizecomparetest.h
minpatternsearch.h
minpatternsearchorder2.h
partialderivative.h
partialderivativetest.cpp
partialderivativetest.h
patternsearch.h
patternsearchD2.h
powell02.h
powell03.h
powelltest.cpp
powelltest.h
prob_func.cpp
prob_func.h
probparab.cpp
probparab.h
probrosenbrock.h
probsysequ01.h

data01.txt
data02.txt
projcompile.txt
unittestsreport.txt

Doxygen

main.cpp
Makefile
exploreh
exploreline
explorequad
explorerand
leastsqrs
lineopgold
lineoptimizergold
lineoptimizergold2
lineoptimizerparabola
linepathD1
minexpdim
minexpdimN
minexpdimtest
minpatternsearch
minpatternsearchorder2
partialderivative
patternsearch
patternsearchD2
powell02
powell03
probparab
probparab2
probparab3
probrosenbrock01
probsysequ01

Intro

This project is a look at minimizing with different algorithms. I have experimented with templated code without virtual functions, but I am now writing something simpler and easier to use.

Minimization of function. Moves to a better solution with each step. ie searches for the local minimum.

I do not know much about optimization and am learning as I go.

TODO

• Pattern search and 2nd order both perform worse than the simple exploratory search. Find out why? Graphically plot a 2D example with methods and their vectors.
• Implement steepest decent with known f(X), f'(X). Compare
• Implement line minimizers.
• Get pattern search and other optimization examples, compare their optimization data with my exploratory and pattern search algorithm implementations.
• Implement a quadratic converging algorithm such as Powell.
• partial derivative explorer
• re-implement exploreline
  suspect hi being changed. Implement reset and have operator ++ try line min in prev direction before an explore move. Like patternsearch implement exp.move after wards. Consider using composition instead. Print hi[] to be sure that there is no corruption.
• kill powell01
• Implement Powell02 with composition
• Partial Derivatives Class Aggressive needed
  When the partial derivative is calculated the two points on either side are calculated for. If these are better than the current minimum implement the move. This makes this not independent because the state itself can change.
• Find out why exploreh fails on Rosenbrock function
• explore operators
  write an explore operator that evaluates until a state change.
  
• Implement an Order Driven explorerh.
  When a variable is successively modified, change the direction of the optimization. For example evaluate N iterations so the operator terminates. Have the first variable evaluated/minimized. Move to the second. If a minimum is found reverse direction and go left.
  Keep minimizing in left direction till no significant change. A measure of when to stop is needed.
  Then go in right direciton minimizing until the is a successful minimization. When there is switch directions and go left.
  ...
• Implement an Order Driven Pattern Search Algorithm
• Implement Pattern Search
• Kill geometriccurve by incorperating the example here
• Redocument the comments in Doxygen style
• Document main so what the test programs do is clearly documented
• Implement 2nd order Pattern Search
• Implement 3rd order Pattern Search
• Test with the Rosenbrock function.
• Partial derivatives.
  f'(0) = [f(h)-f(-h)]/(2h)
  f''(0) = [f'(0.5h)-f'(-0.5h)]/(h)
  f'(0.5) = [f(h)-f(0)]/(h)
  f'(-0.5) = [f(0)-f(-h)]/(h)
  f'''(0) = [f''(0.5h)-f''(-0.5h)]/(h)
  f''(0.5h) = [f'(h)-f'(0)]/(h)
  continue on
  count the number of function evaluations per dimension
  Confirm, discuss with someone