Optimization in Scilab
Manas Das
Department of Electrical Engineering
Indian Institute of Technology Bombay
FOSSEE
21 Feb, 2012
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Todays focus
Scilab is free.
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Todays focus
Scilab is free.
Matrix/loops syntax is similar to Matlab.
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Todays focus
Scilab is free.
Matrix/loops syntax is similar to Matlab.
Scilab provides all basic and many advanced tools.
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Todays focus
Scilab is free.
Matrix/loops syntax is similar to Matlab.
Scilab provides all basic and many advanced tools.
Today: Optimization in Scilab optim command
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Optimization
Minimize/maximize some (more than one) objective function
by varying decision variable
Subject to constraints on functions of decision variables
All engineering domains, economics
Vast area in itself
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Single Objective Optimization Problems
min f (x)
x
where x is n dimensional, f (x) is a scalar, with/without
Bound constraints: xL x xU
Linear/Nonlinear equality constraints: g (x) = b
Linear/Nonlinear inequality constraints: g (x) b
Components of x can be real, integer or binary.
Find Decision Variables: x which minimize Obj: f
Manas Das
Optimization
(1)
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Linear optimization with Linear constraints
Scilab function: karmarkar
Calling Sequence
x1=karmarkar(a,b,c,lb)
a:matrix (n,p)
b:n-vector
c:p-vector
lb:lower bound
Also be solved in scilab using linpro function
linpro is in Quapro toolbox under Optimization in
ATOMS,
It can installed by command atomsInstall(Quapro)
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Linear optimization with Linear constraints
Example:Maximize
3x1 + x2 + 3x3
for
2x1 + x2 + x3 2
x1 + 2x2 + 3x3 5
2x1 + 2x2 + x3 6
x1, x2, x3 0
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Quadratic with linear constraints
Scilab function: qp solve
Calling Sequence
x1=[x [,iact [,iter [,f]]]]=qp solve(Q,p1,C1,b,me)
Q:real positive definite symmetric matrix
(dimension n x n). p- real (column) vector
(dimension n)
c:p - vector
lb:lower bound
me: number of equality constraints
It can also be solved in scilab using quapro function.
quapro is in Quapro toolbox under Optimization in
ATOMS
It can installed by command atomsInstall(quapro)
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Quadratic with linear constraints
Example Minimize
T
2
1 T 2 0
f (x) = 2 x
x+
x
0 2
2
for
x1 + x2 2 + 2
-x1 + x2 2
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Other Optimization Tools in Scilab: I
Native (built-in) functions in Scilab:
fminsearch: Computes unconstrained minimum of a given
function using the Nelder-Mead algorithm (simplex region
search).
Native (built-in function) to Scilab:
Derivative free, relies on search
Basic usage: [xopt,fopt]=fminsearch(costf,x0); costf should
take x as input and return f as output
Try minimizing f (x) = x 2 + 5 using fminsearch
neldermead: Computes minimum of a nonlinear function
subject to bounds and nonlinear constraints
Option for several types of search algorithms
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Unconstrained Non-linear
Scilab function: fminsearch
Calling Sequence
x = fminsearch ( costf , x0 )
costf:The cost function.
x0:The initial guess.
computes the unconstrained minimimum of given
function with the Nelder-Mead algorithm.
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Unconstrained Non-linear
Example: f (x) = x12 + x22 + x1 x2
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Optim command in Scilab
Simplified Call:
[fopt,xopt]=optim(costf,x0)
where
x0: initial guess where minimum of f occurs,
fopt: the optimum (minimum) value,
xopt: where the optimum occurs,
costf : user specified (scilab) function to compute objective
function (f) and/or its gradient (g) depending on the input
integer flag ind as:
[f,g,ind]=costf(x,ind)
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Simple example
f (x) = x 2 + 10
g (x) = 2x
Minimize f (x) starting from x0 = 10.
Write a function (example myfunction1.sci) which takes
x, ind as input and returns f , g , ind as output.
Use optim function to minimize.
Answer: fopt = 10, xopt = 0.
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Another example (Campbell et al., 2010)
f (x1 , x2 , x3 ) = (x1 x3 )2 + 3(x1 + x2 + x3 1)2 + (x1 x3 + 1)2
with gradient
g
f f f
,
,
]
x1 x2 x3
= [2(x1 x3 ) + 6(x1 + x2 + x3 1) + 2(x1 x3 + 1),
= [
6(x1 + x2 + x3 1), 2(x1 x3 ) 2(x1 x3 + 1) + 6(x1 + x2 + x
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Exercise plan
Minimize the function starting with guess values x0 = [1, 1, 1].
Write a function (example myfunction2.sci) which takes
x, ind as input and returns f , g , ind as output.
Use optim function to minimize.
Answer: xopt = [0.14, 0.27, 0.64], fopt = 0.51.
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Other options with optim
Several other options/features with optim
Several optimization algorithms: quasi-newton, conjugate
gradient, etc.
Gradient computation using finite differences: NDcost used
with optim.
Various stopping/diagnostic criteria: Maximum number of
iterations, calls to costf, thresholds, etc.
Use of ind to compute gradient only when required (save
computational effort).
help optim: to know more
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Other Optimization Tools in Scilab: II
Native (built-in) functions in Scilab:
AI based algorithms
Genetic algorithms: single objective optimization,
multiobjective optimization, various implementations
Simulated annealing
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Optimization Toolboxes in Scilab
Can be installed easily (atoms or otherwise) or interfaces to them
quapro: linear, quadratic programming (Q not necessarily
positive definite)
IPopt, fsqp, lp solve, etc.
Check http://atoms.scilab.org/categories/optimization
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optimization Features available in Scilab
Nonlinear optimization with the optim function
Quadratic optimization with the qpsolve function
Nonlinear least-square optimization with the lsqrsolve
function
Semidefinite programming with the semidef function
Genetic algorithms with the optim ga function
Simulated annealing with the optim sa function
Linear matrix inequalities with the lmisolver function
Manas Das
Optimization
Optim in Scilab
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
Missing optimization features in Scilab
A list of features which are not available in Scilab,
but are available in toolboxes.
Integer parameter with linear objective solver and sparse
matrices : currently available in LPSOLVE toolbox, based on
the simplex method,
Linear objective with sparse matrices : currently available in
LIPSOL, based on interior points method,
Nonlinear objective and non linear constraints : currently
available in interface to IPOPT toolbox, based on interior
point methods,
Nonlinear objective and non linear constraints : currently
available in interface to CONMIN toolbox, based on method
of feasible directions,
Manas Das
Optimization
Outline
Scilab
Optimization
Other Optimization Tools in Scilab
Optim in Scilab
References
S. L. Campbell, J.P. Chancelier and R. Nikoukhah, Modeling
and Simulation in Scilab/Scicos, Springer, 2006.
Thank You
Manas Das
Optimization
