expressionMutation: Random mutation of functions and expressions
In rgp: R genetic programming framework

Description Usage Arguments Details Value

RGP implements two sets of mutation operators. The first set is inspired by classical GP systems. Mutation strength is controlled by giving mutation probabilities: mutateFunc mutates a function f by recursively replacing inner function labels in f with probability mutatefuncprob. mutateSubtree mutates a function by recursively replacing inner nodes with newly grown subtrees of maximum depth maxsubtreedepth. mutateNumericConst mutates a function by perturbing each numeric (double) constant c with probability mutateconstprob by setting c := c + rnorm(1, mean = mu, sd = sigma). Note that constants of other typed than double (e.g integers) are not affected.

mutateFunc(func, funcset, mutatefuncprob = 0.1,
  breedingFitness = function(individual) TRUE, breedingTries = 50)

mutateSubtree(func, funcset, inset, conset, mutatesubtreeprob = 0.1,
  maxsubtreedepth = 5, breedingFitness = function(individual) TRUE,
  breedingTries = 50)

mutateNumericConst(func, mutateconstprob = 0.1,
  breedingFitness = function(individual) TRUE, breedingTries = 50, mu = 0,
  sigma = 1)

mutateFuncTyped(func, funcset, mutatefuncprob = 0.1,
  breedingFitness = function(individual) TRUE, breedingTries = 50)

mutateSubtreeTyped(func, funcset, inset, conset, mutatesubtreeprob = 0.1,
  maxsubtreedepth = 5, breedingFitness = function(individual) TRUE,
  breedingTries = 50)

mutateNumericConstTyped(func, mutateconstprob = 0.1,
  breedingFitness = function(individual) TRUE, breedingTries = 50)

mutateChangeLabel(func, funcset, inset, conset, strength = 1,
  breedingFitness = function(individual) TRUE, breedingTries = 50)

mutateInsertSubtree(func, funcset, inset, conset, strength = 1,
  subtreeDepth = 2, breedingFitness = function(individual) TRUE,
  breedingTries = 50)

mutateDeleteSubtree(func, funcset, inset, conset, strength = 1,
  subtreeDepth = 2, constprob = 0.2,
  breedingFitness = function(individual) TRUE, breedingTries = 50)

mutateChangeDeleteInsert(func, funcset, inset, conset, strength = 1,
  subtreeDepth = 2, constprob = 0.2, iterations = 1,
  changeProbability = 1/3, deleteProbability = 1/3,
  insertProbability = 1/3, breedingFitness = function(individual) TRUE,
  breedingTries = 50)

mutateDeleteInsert(func, funcset, inset, conset, strength = 1,
  subtreeDepth = 2, constprob = 0.2, iterations = 1,
  deleteProbability = 0.5, insertProbability = 0.5,
  breedingFitness = function(individual) TRUE, breedingTries = 50)

mutateFuncFast(funcbody, funcset, mutatefuncprob = 0.1)

mutateSubtreeFast(funcbody, funcset, inset, constmin, constmax, insertprob,
  deleteprob, subtreeprob, constprob, maxsubtreedepth)

mutateNumericConstFast(funcbody, mutateconstprob = 0.1, mu = 0, sigma = 1)

`func`	The function to mutate randomly.
`funcbody`	The function body to mutate randomly, obtain it via `body(func)`.
`funcset`	The function set.
`inset`	The set of input variables.
`conset`	The set of constant factories.
`mutatefuncprob`	The probability of trying to replace an inner function at each node.
`mutatesubtreeprob`	The probability of replacing a subtree with a newly grown subtree at each node.
`maxsubtreedepth`	The maximum depth of newly grown subtrees.
`mutateconstprob`	The probability of mutating a constant by adding `rnorm(1)` to it.
`strength`	The number of individual point mutations (changes, insertions, deletions) to perform.
`subtreeDepth`	The depth of the subtrees to insert or delete.
`constprob`	The probability of creating a constant versus an input variable.
`insertprob`	The probability to insert a subtree.
`deleteprob`	The probability to insert a subtree.
`constmin`	The lower limit for numeric constants.
`constmax`	The upper limit for numeric onstants.
`mu`	The normal distribution mean for random numeric constant mutation.
`sigma`	The normal distribution standard deviation for random numeric constant mutation.
`subtreeprob`	The probability of creating a subtree instead of a leaf in the random subtree generator function.
`iterations`	The number of times to apply a mutation operator to a GP individual. This can be used as a generic way of controling the strength of the genotypic effect of mutation.
`changeProbability`	The probability for selecting the `mutateChangeLabel` operator.
`deleteProbability`	The probability for selecting the `mutateDeleteSubtree` operator.
`insertProbability`	The probability for selecting the `mutateInsertSubtree` operator.
`breedingFitness`	A breeding function. See the documentation for `geneticProgramming` for details.
`breedingTries`	The number of breeding steps.

mutateFuncTyped, mutateSubtreeTyped, and mutateNumericConstTyped are variants of the above functions that only create well-typed result expressions.

mutateFuncFast, mutateSubtreeFast, mutateNumericConstFast are variants of the above untyped mutation function implemented in C. They offer a considerably faster execution speed for the price of limited flexibility. These variants take function bodies as arguments (obtain these via R's body function) and return function bodies as results. To turn a function body into a function, use RGP's makeClosure tool function.

The second set of mutation operators features a more orthogonal design, with each individual operator having a only a small effect on the genotype. Mutation strength is controlled by the integral strength parameter. mutateChangeLabel Selects a node (inner node or leaf) by uniform random sampling and replaces the label of this node by a new label of matching type. mutateInsertSubtree Selects a leaf by uniform random sampling and replaces it with a matching subtree of the exact depth of subtreeDepth. mutateDeleteSubtree Selects a subree of the exact depth of subtreeDepth by uniform random sampling and replaces it with a matching leaf. mutateChangeDeleteInsert Either applies mutateChangeLabel, mutateInsertSubtree, or mutateDeleteSubtree. The probability weights for selecting an operator can be supplied via the ...Probability arguments (probability weights are normalized to a sum of 1). mutateDeleteInsert Either applies mutateDeleteSubtree or mutateInsertSubtree. The probability weights for selecting an operator can be supplied via the ...Probability arguments (probability weights are normalized to a sum of 1). The above functions automatically create well-typed result expressions when used in a strongly typed GP run.