discretize {arules}  R Documentation 
This function implements several basic unsupervised methods to convert a continuous variable into a categorical variable (factor) using different binning strategies. For convenience, a whole data.frame can be discretized (i.e., all numeric columns are discretized).
discretize(
x,
method = "frequency",
breaks = 3,
labels = NULL,
include.lowest = TRUE,
right = FALSE,
dig.lab = 3,
ordered_result = FALSE,
infinity = FALSE,
onlycuts = FALSE,
categories = NULL,
...
)
discretizeDF(df, methods = NULL, default = NULL)
x 
a numeric vector (continuous variable). 
method 
discretization method. Available are: 
breaks , categories 
either number of categories or a vector with boundaries for
discretization (all values outside the boundaries will be set to NA).

labels 
character vector; labels for the levels of the resulting
category. By default, labels are constructed using "(a,b]" interval
notation. If 
include.lowest 
logical; should the first interval be closed to the left? 
right 
logical; should the intervals be closed on the right (and open on the left) or vice versa? 
dig.lab 
integer; number of digits used to create labels. 
ordered_result 
logical; return a ordered factor? 
infinity 
logical; should the first/last break boundary changed to +/Inf? 
onlycuts 
logical; return only computed interval boundaries? 
... 
for method "cluster" further arguments are passed on to

df 
data.frame; each numeric column in the data.frame is discretized. 
methods 
named list of lists or a data.frame; the named list contains
lists of discretization parameters (see parameters of 
default 
named list; parameters for 
Discretize calculates breaks between intervals using various methods and
then uses base::cut()
to convert the numeric values into intervals
represented as a factor.
Discretization may fail for several reasons. Some reasons are
A variable contains only a single value. In this case, the variable should be dropped or directly converted into a factor with a single level (see factor).
Some calculated breaks are not unique. This can happen for method frequency with very skewed data (e.g., a large portion of the values is 0). In this case, nonunique breaks are dropped with a warning. It would be probably better to look at the histogram of the data and decide on breaks for the method fixed.
discretize
only implements unsupervised discretization. See
arulesCBA::discretizeDF.supervised()
in package arulesCBA
for supervised discretization.
discretizeDF()
applies discretization to each numeric column.
Individual discretization parameters can be specified in the form:
methods = list(column_name1 = list(method = ,...), column_name2 = list(...))
.
If no discretization method is specified for a column, then the
discretization in default is applied (NULL
invokes the default method
in discretize()
). The special method "none"
can be specified
to suppress discretization for a column.
discretize()
returns a factor representing the
categorized continuous variable with
attribute "discretized:breaks"
indicating the used breaks or and
"discretized:method"
giving the used method. If onlycuts = TRUE
is used, a vector with the calculated interval boundaries is returned.
discretizeDF()
returns a discretized data.frame.
Michael Hahsler
base::cut()
,
arulesCBA::discretizeDF.supervised()
.
Other preprocessing:
hierarchy
,
itemCoding
,
merge()
,
sample()
data(iris)
x < iris[,1]
### look at the distribution before discretizing
hist(x, breaks = 20, main = "Data")
def.par < par(no.readonly = TRUE) # save default
layout(mat = rbind(1:2,3:4))
### convert continuous variables into categories (there are 3 types of flowers)
### the default method is equal frequency
table(discretize(x, breaks = 3))
hist(x, breaks = 20, main = "Equal Frequency")
abline(v = discretize(x, breaks = 3,
onlycuts = TRUE), col = "red")
# Note: the frequencies are not exactly equal because of ties in the data
### equal interval width
table(discretize(x, method = "interval", breaks = 3))
hist(x, breaks = 20, main = "Equal Interval length")
abline(v = discretize(x, method = "interval", breaks = 3,
onlycuts = TRUE), col = "red")
### kmeans clustering
table(discretize(x, method = "cluster", breaks = 3))
hist(x, breaks = 20, main = "KMeans")
abline(v = discretize(x, method = "cluster", breaks = 3,
onlycuts = TRUE), col = "red")
### userspecified (with labels)
table(discretize(x, method = "fixed", breaks = c(Inf, 6, Inf),
labels = c("small", "large")))
hist(x, breaks = 20, main = "Fixed")
abline(v = discretize(x, method = "fixed", breaks = c(Inf, 6, Inf),
onlycuts = TRUE), col = "red")
par(def.par) # reset to default
### prepare the iris data set for association rule mining
### use default discretization
irisDisc < discretizeDF(iris)
head(irisDisc)
### discretize all numeric columns differently
irisDisc < discretizeDF(iris, default = list(method = "interval", breaks = 2,
labels = c("small", "large")))
head(irisDisc)
### specify discretization for the petal columns and don't discretize the others
irisDisc < discretizeDF(iris, methods = list(
Petal.Length = list(method = "frequency", breaks = 3,
labels = c("short", "medium", "long")),
Petal.Width = list(method = "frequency", breaks = 2,
labels = c("narrow", "wide"))
),
default = list(method = "none")
)
head(irisDisc)
### discretize new data using the same discretization scheme as the
### data.frame supplied in methods. Note: NAs may occure if a new
### value falls outside the range of values observed in the
### originally discretized table (use argument infinity = TRUE in
### discretize to prevent this case.)
discretizeDF(iris[sample(1:nrow(iris), 5),], methods = irisDisc)