mgcv (version 1.9-0)

uniquecombs: find the unique rows in a matrix

Description

This routine returns a matrix or data frame containing all the unique rows of the matrix or data frame supplied as its argument. That is, all the duplicate rows are stripped out. Note that the ordering of the rows on exit need not be the same as on entry. It also returns an index attribute for relating the result back to the original matrix.

Usage

uniquecombs(x,ordered=FALSE)

Value

A matrix or data frame consisting of the unique rows of x (in arbitrary order).

The matrix or data frame has an "index" attribute. index[i] gives the row of the returned matrix that contains row i of the original matrix.

Arguments

x

is an R matrix (numeric), or data frame.

ordered

set to TRUE to have the rows of the returned object in the same order regardless of input ordering.

Author

Simon N. Wood with thanks to Jonathan Rougier

WARNINGS

If a dataframe contains variables of a type other than numeric, logical, factor or character, which either have no as.character method, or whose as.character method is a many to one mapping, then the routine is likely to fail.

If the character representation of a dataframe variable (other than of class factor of character) contains * then in principle the method could fail (but with a warning).

Details

Models with more parameters than unique combinations of covariates are not identifiable. This routine provides a means of evaluating the number of unique combinations of covariates in a model.

When x has only one column then the routine uses unique and match to get the index. When there are multiple columns then it uses paste0 to produce labels for each row, which should be unique if the row is unique. Then unique and match can be used as in the single column case. Obviously the pasting is inefficient, but still quicker for large n than the C based code that used to be called by this routine, which had O(nlog(n)) cost. In principle a hash table based solution in C would be only O(n) and much quicker in the multicolumn case.

unique and duplicated, can be used in place of this, if the full index is not needed. Relative performance is variable.

If x is not a matrix or data frame on entry then an attempt is made to coerce it to a data frame.

See Also

unique, duplicated, match.

Examples

require(mgcv)

## matrix example...
X <- matrix(c(1,2,3,1,2,3,4,5,6,1,3,2,4,5,6,1,1,1),6,3,byrow=TRUE)
print(X)
Xu <- uniquecombs(X);Xu
ind <- attr(Xu,"index")
## find the value for row 3 of the original from Xu
Xu[ind[3],];X[3,]

## same with fixed output ordering
Xu <- uniquecombs(X,TRUE);Xu
ind <- attr(Xu,"index")
## find the value for row 3 of the original from Xu
Xu[ind[3],];X[3,]


## data frame example...
df <- data.frame(f=factor(c("er",3,"b","er",3,3,1,2,"b")),
      x=c(.5,1,1.4,.5,1,.6,4,3,1.7),
      bb = c(rep(TRUE,5),rep(FALSE,4)),
      fred = c("foo","a","b","foo","a","vf","er","r","g"),
      stringsAsFactors=FALSE)
uniquecombs(df)