FinMath.MachineLearning.EvolutionaryAlgorithms Namespace

Basic class for some EvolutionaryAlgorithms optimizers.

The objective change stop criterion.

Implements the CMSA. The algorithm is described in H.G.Beyer, B.Sendhoff (2008). Covariance Matrix Adaptation Revisited: The CMSA Evolution Strategy In Proceedings of the Tenth International Conference on Parallel Problem Solving from Nature (PPSN X), pp. 123-132, LNCS, Springer-Verlag Also see http://en.wikipedia.org/wiki/CMA-ES

This class implements the noisy cross entropy method as described in the following article. Christophe Thiery, Bruno Scherrer.Improvements on Learning Tetris with Cross Entropy. International Computer Games Association Journal, ICGA, 2009, 32. The algorithm aims to minimize an objective function through stochastic search. It works iteratively until a certain stopping criteria is met. At each iteration, it samples a number of vectors from a Gaussian distribution and evaluates each of these against the supplied objective function. Based on the return value from the objective function, a subset of the the best ranked vectors are chosen to update the search parameters of the next generation. The mean of the Gaussian distribution is set to the centroid of the best ranked vectors, and the variance is set to the variance of the best ranked vectors in each individual dimension.

Algorithm logic class

The Brute-force search algorithm on parameters grid.

The random points running over algorithm.

Roulette selection chooses parents by simulating a roulette wheel, in which the area of the section of the wheel corresponding to an individual is proportional to the individual's scaled fitness. The algorithm uses a random number to select one of the sections with a probability equal to its area.

The Search algorithm base class.

The categorical factor class.

Continuous factor description class.

The discrete continuous factor class.

Interface for selection algorithm.

The Nelder-Mead Simplex Downhill Method is a deterministic direct search method. It is known to perform quite well in low dimensions, at least for local search. http://en.wikipedia.org/wiki/Nelder%E2%80%93Mead_method

Stochastic uniform lays out a line in which each parent corresponds to a section of the line of length proportional to its scaled value. The algorithm moves along the line in steps of equal size. At each step, the algorithm allocates a parent from the section it lands on.

Tournament selection chooses each parent by choosing tournamentSize individual at random and then choosing the best individual out of that set to be a parent.

Uniform selection chooses each individual with equal probability.

Implements the VD-CMA-ES Algorithm The VD-CMA-ES implements a restricted form of the CMA-ES where the covariance matrix is restricted to be (D+vv^T) where D is a diagonal matrix and v a single vector. Therefore this variant is capable of large-scale optimization For more reference, see the paper Akimoto, Y., A. Auger, and N. Hansen (2014). Comparison-Based Natural Gradient Optimization in High Dimension. To appear in Genetic and Evolutionary Computation Conference (GECCO 2014), Proceedings, ACM The implementation differs from the paper to be closer to the reference implementation and to have better numerical accuracy.

The countable factor description base class.

The factor description interface.

The delegate to the function that computes the value of the objective function.

The delegate to the function that determines stopping criterion.

Creates a new individual.

Crossovers two individuals and produces one child.

Crossovers two individuals and produces two children.

Elite delegate.

Exit condition delegate.

Mutates individual.

Block processing objective delegate.

Objective delegate. The optimization process try minimize objective value.

Type of direct search method.

Type of fitness scaling used by selection

Selection probability function distribution type