cspade {arulesSequences}

Mining Associations with cSPADE

Description

Mining frequent sequential patterns with the cSPADE algorithm. This algorithm utilizes temporal joins along with efficient lattice search techniques and provides for timing constraints.

Usage

cspade(data, parameter = NULL, control = NULL, tmpdir = tempdir())

Arguments

data

an object of class transactions with temporal information.

parameter

an object of class SPparameter or a named list with corresponding components.

control

an object of class SPcontrol or a named list with corresponding components.

tmpdir

a non-empty character vector giving the directory name where temporary files are written.

Details

Interfaces the command-line tools for preprocessing and mining frequent sequences with the cSPADE algorithm by M. Zaki via a proper chain of system calls.

The temporal information is taken from components sequenceID (sequence or customer identifier) and eventID (event identifier) of transactionInfo. Note that integer identifiers must be positive and that transactions must be ordered by sequenceID and eventID.

Class information (on sequences or customers) is taken from component classID, if available.

The amount of disk space used by temporary files is reported in verbose mode (see class SPcontrol).

If specified timeout is passed to system2 (see details there and class SPcontrol).

Value

Returns an object of class sequences.

Warning

The implementation of the maxwin constraint in the command-line tools seems to be broken. To avoid confusion it is disabled with a warning.

Note

Temporary files may not be deleted until the end of the R session if the call is interrupted. Use timeouts to avoid this problem.

The current working directory (see getwd) must be writable.

Author(s)

Christian Buchta, Michael Hahsler

References

M. J. Zaki. (2001). SPADE: An Efficient Algorithm for Mining Frequent Sequences. Machine Learning Journal, 42, 31–60.

See Also

Class transactions, sequences, SPparameter, SPcontrol, method ruleInduction, support, function read_baskets.

Examples

## use example data from paper
data(zaki)
## get support bearings
s0 <- cspade(zaki, parameter = list(support = 0,
                                    maxsize = 1, maxlen = 1),
                   control   = list(verbose = TRUE))
as(s0, "data.frame")
## mine frequent sequences
s1 <- cspade(zaki, parameter = list(support = 0.4), 
		   control   = list(verbose = TRUE, tidLists = TRUE))
summary(s1)
as(s1, "data.frame")

##
summary(tidLists(s1))
transactionInfo(tidLists(s1))

## use timing constraint
s2 <- cspade(zaki, parameter = list(support = 0.4, maxgap = 5))
as(s2, "data.frame")

## use classification
t <- zaki
transactionInfo(t)$classID <-
    as.integer(transactionInfo(t)$sequenceID) %% 2 + 1L
s3 <- cspade(t, parameter = list(support = 0.4, maxgap = 5))
as(s3, "data.frame")

## replace timestamps
t <- zaki
transactionInfo(t)$eventID <-
    unlist(tapply(seq(t), transactionInfo(t)$sequenceID,
	function(x) x - min(x) + 1), use.names = FALSE)
as(t, "data.frame")
s4 <- cspade(t, parameter = list(support = 0.4))
s4
identical(as(s1, "data.frame"), as(s4, "data.frame"))

## work around
s5 <- cspade(zaki, parameter = list(support = .25, maxgap = 5))
length(s5)
k <- support(s5, zaki, control   = list(verbose = TRUE,
                       parameter = list(maxwin = 5)))
table(size(s5[k == 0]))

## Not run: 
## use generated data
t <- read_baskets(con  = system.file("misc", "test.txt", package =
				      "arulesSequences"),
		  info = c("sequenceID", "eventID", "SIZE"))
summary(t)
## use low support
s6 <- cspade(t, parameter = list(support = 0.0133), 
		control   = list(verbose = TRUE, timeout = 15))
summary(s6)

## check
k <- support(s6, t, control = list(verbose = TRUE))
table(size(s6), sign(quality(s6)$support -k))

## use low confidence
r6 <- ruleInduction(s6, confidence = .5,
			control    = list(verbose = TRUE))
summary(r6)

## End(Not run)