reduce {lfl}

Reduce the size of rule base

Description

From given rule base, select such set of rules that influence mostly the rule base coverage of the input data.

Usage

reduce(
  x,
  rules,
  ratio,
  tnorm = c("goedel", "goguen", "lukasiewicz"),
  tconorm = c("goedel", "goguen", "lukasiewicz"),
  numThreads = 1
)

Arguments

x

Data for the rules to be evaluated on. Could be either a numeric matrix or numeric vector. If matrix is given then the rules are evaluated on rows. Each value of the vector or column of the matrix represents a predicate - it's numeric value represents the truth values (values in the interval [0, 1]).

rules

Either an object of class "farules" or list of character vectors where each vector is a rule with consequent being the first element of the vector. Elements of the vectors (predicate names) must correspond to the x's names (of columns if x is a matrix).

ratio

A percentage of rule base coverage that must be preserved. It must be a value within the [0, 1] interval. Value of 1 means that the rule base coverage of the result must be the same as coverage of input rules. A sensible value is e.g. 0.9.

tnorm

Which t-norm to use as a conjunction of antecedents. The default is "goedel".

tconorm

Which t-norm to use as a disjunction, i.e. to combine multiple antecedents to get coverage of the rule base. The default is "goedel".

numThreads

How many threads to use for computation. Value higher than 1 causes that the algorithm runs in several parallel threads (using the OpenMP library).

Details

From a given rulebase, a rule with greatest coverage is selected. After that, additional rules are selected that increase the rule base coverage the most. Addition stops after the coverage exceeds original coverage * ratio.

Note that the size of the resulting rule base is not necessarily minimal because the algorithm does not search all possible combination of rules. It only finds a local minimum of rule base size.

Value

Function returns an instance of class farules() or a list depending on the type of the rules argument.

Author(s)

Michal Burda

References

M. Burda, M. Štěpnička, Reduction of Fuzzy Rule Bases Driven by the Coverage of Training Data, in: Proc. 16th World Congress of the International Fuzzy Systems Association and 9th Conference of the European Society for Fuzzy Logic and Technology (IFSA-EUSFLAT 2015), Advances in Intelligent Systems Research, Atlantic Press, Gijon, 2015.

See Also

rbcoverage(), farules()