The function `smco`

is use to optimize non-linear
complex functions based on three simple ideas: first, the
sampling of each component of the solution vector, one at a time,
based on a truncated normal distribution; second, the evolution
of the standard deviation of the sampling distribution in each iteration,
as a mechanism of self-adaptation; and third,
the restart of the algorithm for escaping of local optima.

1 2 3 4 |

`par` |
Initial values for the parameters to be optimized over. When it is 'NULL' the user must be specify the value for the 'N' parameter and the algorithm it is initializated in a random point. |

`fn` |
A function to be minimized (or maximized), with first argument the vector of parameters over which minimization is to take place. It should return a scalar result. |

`gr` |
A function to return the gradient for the '"L-BFGS-B"' method when
the option |

`...` |
Further arguments to be passed to 'fn' and 'gr'. |

`N` |
The number of dimensions of the vector of solution variables |

`LB, UB` |
Lower and upper bounds on the variables. |

`maxiter` |
The maximum number of iterations. The number of function calls to '"L-BFGS-B"' implementation is not taking into count. |

`Co, Cmin, Cmax` |
Initial, minimum and maximum values for the standard deviation of the truncated normal distribution |

`trc` |
Logical. When it is 'TRUE' the progress of the optimization algorithm is reported. |

`lambda` |
Restarting parameter. When the algorithm reaches 'lambda' * 'N' iterations without found a new local optimum the restart ocurrs. |

`useBFGS` |
Logical. When it value is 'TRUE', first, a new random
value for the current coordinate is generated using a truncated
normal distribution; second, the one-dimensional optimization of
the value of the current coordinate is made using the
'"L-BFGS-B"' method implemented in the |

`control` |
A list of control parameters for the '"L-BFGS-B"' implemented
in the |

`hessian` |
A control parameter for the '"L-BFGS-B"' implemented
in the |

The algorithm tries to improve the current solution by changing the
i-th coordinate by a random value generated using a truncated normal
distribution. Only a random value for each coordinate is generated
each time. When the parameter 'useBFGS' is 'TRUE', the new random
value is used as the initial point for the one-dimensional
optimization of the i-th coordinate using '"L-BFGS-B"' method
implemented in the `optim`

function.

The standard deviation of the truncated random deviation changes in
each iteration in a random way. The values of the standard deviation
are bounded to the interval defined by 'Cmin' and 'Cmax'.

When the algorithm reaches 'lambda' * 'N' iterations without found
a new local optimum the restart ocurrs. For this, a new random
current solution is generated.

A list with components:

`par` |
The best set of parameters found. |

`value` |
The value of 'fn' corresponding to 'par'. |

`f.opt, f.min, f.curr` |
Vectors of size 'maxiter' containing the values of the global optimum, local optimum and current value of the function 'fn' for the current iteration. |

`call` |
The call to the function. |

Prof. Juan D. Velasquez, Ph.D.

Grupo de Computacion Aplicada

Univesidad Nacional de Colombia

jdvelasq@unal.edu.co

Velasquez, J. D. (2011). *A Simple Monte Carlo optimizer based on
Adaptive Coordinate Sampling*. Submitted to Operation Research Letters.

`optim`

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 | ```
SphereModel.fcn <-
function( x ) {
return(sum(x ^ 2))
}
f = SphereModel.fcn;
ndim = 10;
LB = rep( -600.000, ndim);
UB = rep( 600.000, ndim);
maxiter = 100;
s = smco(par = NULL, fn = SphereModel.fcn, N = ndim, LB = LB,
UB = UB, maxiter = maxiter, Co = 0.01, Cmin = 0.0001,
Cmax = 0.5, trc = TRUE, lambda = 20,
useBFGS = TRUE, control = list(maxit = 10))
``` |

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.