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 integer
s) are not affected.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49  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 
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 
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 
deleteProbability 
The probability for selecting the 
insertProbability 
The probability for selecting the 
breedingFitness 
A breeding function. See the documentation for

breedingTries 
The number of breeding steps. 
mutateFuncTyped
, mutateSubtreeTyped
, and mutateNumericConstTyped
are
variants of the above functions that only create welltyped 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 welltyped result expressions when used in a strongly
typed GP run.
All RGP mutation operators have the S3 class c("mutationOperator", "function")
.
The randomly mutated function.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.