Grid mutation

This mutation operator is designed to introduce a partial grid structure into to point of a given Euclidean problem instance. This is achieved by generating a random box of width and height \(w, h \in [box.min, box.max]\) and placing it at random within the boundaries \([0, 1]^2\). All points \(Q \subseteq P\) inside the box are affected by the mutation. Point set \(Q\) is replaced by a quadratic grid of points \(Q'\) with \(g = \lfloor \sqrt{|Q|} \rfloor\) rows and columns respectively. Note that if \(g^2 < |Q|\) we ignore \(|Q| - g\) random points of \(Q\), i.e., these points are not touched. Subsequent, optional steps involve rotation and noise addition: with probability \(p.rot\) \(Q'\) is rotated by a random angle \(alpha \in [0, \pi/2]\) and with probability p.jitter all points in \(Q\) are perturbed by additive Gaussian noise with mean \((0,0)\) and standard deviation jitter.sd in each dimension.

doGridMutation(coords, box.min = 0.1, box.max = 0.3, p.rot = 0,
  p.jitter = 0, jitter.sd = 0, ...)

Arguments

coords	[`matrix`] An n times 2 matrix of point coordinates in \([0, 1]^2\).
box.min	[`numeric(1)`] Minimum for sampled box width and height respectively. Default is 0.1.
box.max	[`numeric(1)`] Maximum for sampled box width and height respectively. Default is 0.3.
p.rot	[`numeric(1)`] Probability of rotation, i.e, that mutated points are subject to rotation. Default is 0, i.e., no rotation.
p.jitter	[`numeric(1)`] Probability to add Gaussian noise to mutated points. Default is 0, i.e., no Gaussian noise at all.
jitter.sd	[`numeric(1)`] Standard deviation for Gaussian noise. Defaults to 0.1. Note that this parameter has an effect only if `p.jitter` is greater zero.
...	[any] Currently not used.

Value

[matrix] Mutated coordinates.

Arguments

Value

See also