The Shifted Hammersley Sampling algorithm is used for sample generation by all methods
except NLPQL. The conventional Hammersley sampling algorithm is a quasi-random number
generator, which has very low discrepancy and is used for quasi-Monte Carlo simulations. A
low-discrepancy sequence is defined as a sequence of points that approximate the
equidistribution in a multi-dimensional cube in an optimal way. This means that the design
space is populated almost uniformly by these sequences and that dimensionality is not a
problem because of the inherent properties of Monte Carlo sampling. The number of points
does not increase exponentially with an increase in the number of input parameters. The
conventional Hammersley sampling algorithm is constructed by using the radical inverse
function. Any integer can be represented as a sequence of digits
by the following equation:
(10) |
For example, consider the integer 687459, which can be represented this way
as, and so on. Because this integer is represented with radix 10, you can
write it as
and so on. In general, for a radix
representation, the equation is:
(11) |
The inverse radical function is defined as the function that generates a fraction in (0, 1) by reversing the order of the digits in Equation 11 about the decimal point, as shown.
(12) |
Thus, for a -dimensional search space, the Hammersley points are given by the
following expression:
(13) |
where indicates the sample points. Now, from the plot of these points, it is
seen that the first row (corresponding to the first sample point) of the Hammersley matrix
is zero and the last row is not 1. This implies that, for the
-dimensional hypercube, the Hammersley sampler generates a block of
points that are skewed more toward the origin of the cube and away from the far edges and
faces. To compensate for this bias, a point-shifting process is proposed that shifts all
Hammersley points by the amount that follows:
(14) |
This moves the point set more toward the center of the search space and avoids unnecessary bias. Thus, the initial population always provides unbiased, low-discrepancy coverage of the search space.