discretize: Discretization of numeric attributes

View source: R/Rinterface.R

discretizeR Documentation

Discretization of numeric attributes


The method discretize returns discretization bounds for numeric attributes and two auxiliary functions. Discretization can be obtained with one of the three discretization methods: greedy search using given feature evaluation heuristics, equal width of intervals, or equal number of instances in each interval. The attributes and target variable are specified using formula interface, target variable name or index. Feature evaluation algorithms available for classification problems are various variants of Relief and ReliefF algorithms, gain ratio, gini-index, MDL, DKM, information gain, etc. For regression problems there are RREliefF, MSEofMean, MSEofModel, MAEofMode, etc.


 discretize(formula, data, method=c("greedy", "equalFrequency", "equalWidth"), 
            estimator, discretizationLookahead=3, discretizationSample=0, 
            maxBins=0, equalDiscBins=4, ...)
 applyDiscretization(data, boundsList, noDecimalsInValueName=2)
 intervalMidPoint(data, boundsList, 
                  midPointMethod=c("equalFrequency", "equalWidth")) 



Either a formula specifying the attributes to be evaluated and the target variable, or a name of target variable, or an index of target variable.


Data frame with data.


Three discretization methods are available. With method="greedy" greedy search using given feature evaluation heuristics is selected, while "equalFrequency" and "equalWidth" select equal frequency (the same number of instances in each interval) and equal width discretization, respectively.


The name of the evaluation method.


Discretization is performed with a greedy algorithm which adds a new boundary, until there is no improvement in evaluation function for discretizationLookahead number of times (0=try all possibilities). Candidate boundaries are chosen from a random sample of boundaries, whose size is discretizationSample.


Maximal number of points to try discretization (0=all sensible). Binarization of multivalued discrete features with k values is performed exhaustively, if 2^k - 1 is at most discretizationSample. Otherwise binarization is done greedily starting from the best separation of a single value. For ReliefF-type measures, binarization of numeric features is performed with discretizationSample randomly chosen splits. For other measures, the split is searched exhaustively among all possible splits.


The maximal number of discrete bins for numeric attributes used for greedy discretization (0=don't care). This shall be an integer vector of length equal to the number of numeric attributes or an integer which applies to all numeric attributes. The default value of 0 means that the number of bins will be determined greedily taking into account discretizationLookahead.



The number of bins used in equal frequency and equal width discretization. This shall be an integer vector of length equal to the number of numeric attributes or an integer which applies to all numeric attributes. The default value is 4.


Additional options used by specific evaluation methods as described in helpCore.


A list of numeric bounds which is applied to numeric attributes in data to produce discrete attributes of type factor. Numeric bounds can be obtained by calling discretize function.



With how many decimal places will the numeric feature values be presented in description (i.e., levels) of feature values. The default value is 2, but will be increased if this is necessary to avoid the same description of feature values.


Two methods to determine the middle points of discretization intervals are available. The "equalFrequency" method select the middle point so that each half-interval contains equal number of instances. The "equalWidth" methods sets middle point to be equally distant from the boundaries.


In method discretize the parameter formula can be interpreted in three ways, where the formula interface is the most elegant one, but inefficient and inappropriate for large data sets. See CoreModel for details.

The estimator parameter selects the evaluation heuristics. For classification problem it must be one of the names returned by infoCore(what="attrEval") and for regression problem it must be one of the names returned by infoCore(what="attrEvalReg"). For details see their description in attrEval.

If the number of supplied vector in maxBins and equalDiscBins is shorter than the number of numeric attributes, the vector is coerced to the required length.

There are some additional parameters ... available which are used by specific evaluation heuristics. Their list and short description is available by calling helpCore. See Section on attribute evaluation.

The function applyDiscretization takes the discretization bounds obtain with function discretize and transforms numeric features in a data set into discrete features.

The function intervalMidPoint takes discretization bounds provided by function discretize and returns middle points of discretization intervals for numeric attributes. The middle points are computed from the data; for lowest/highest interval the minimum/maximum of the values in the data for particular attribute are implicitly taken as an additional left/right boundary point.


The method discretize returns a list of discretization bounds for numeric attributes. One component of a list contains bounds for one attribute. If an attribute has all values equal, value NA is returned. If an attribute has all values equal to NA, it is skipped in the returned list.

The function applyDiscretization returns a data set where all numeric attributes are replaced with their discrete versions.

The function intervalMidPoint returns a list of vectors where each vector contains middle point of discretized intevals.


Marko Robnik-Sikonja


Marko Robnik-Sikonja, Igor Kononenko: Theoretical and Empirical Analysis of ReliefF and RReliefF. Machine Learning Journal, 53:23-69, 2003

Marko Robnik-Sikonja, Igor Kononenko: Discretization of continuous attributes using ReliefF. Proceedings of ERK'95 , Portoroz, Slovenia, 1995.

Some of these references are available also from http://lkm.fri.uni-lj.si/rmarko/papers/

See Also

CORElearn, CoreModel, attrEval, helpCore, infoCore.


# use iris data
# run method using estimator ReliefF with exponential rank distance  
discBounds <- discretize(Species ~ ., iris, method="greedy", 
discreteIris <- applyDiscretization(iris, discBounds)
prototypePoints <- intervalMidPoint(iris, discBounds, 

regData <- regDataGen(200)
discretize(response ~ ., regData, method="greedy", estimator="RReliefFequalK", 
# print all available estimators

CORElearn documentation built on Nov. 18, 2022, 5:08 p.m.