Convert vector to a onehot representation (binary class matrix)

Convert vector to a onehot representation (binary class matrix)

onehot_with_decoder(y, order = NULL, named = TRUE)

onehot(y, order = NULL, named = TRUE)

decode_onehot(
  Y,
  classes = colnames(Y),
  n_classes = ncol(Y) %||% length(classes)
)

onehot_decoder(Y, classes = colnames(Y), n_classes = length(classes))

Arguments

y	character, factor, or numeric vector
order	NULL, FALSE, or a character vector. If NULL (the default), then levels are sorted with sort(). If FALSE, then levels are taken in order of their first appearance in y. If a character vector, then order must contain all levels found in y.
named	if the returned matrix should have column names
Y	a matrix, as returned by onehot() or similar.
classes	A character vector of class names in the order corresponding to Y's onehot encoding. Typically, colnames(Y). if NULL, then the decoder returns the column number.
n_classes	The total number of classes expected in Y. Used for input checking in the returned decoder, also, to reconstruct the correct dimensions if the passed in Y is missing dim() attributes.

Value

A binary class matrix

See also

keras::to_categorical

Examples

if(require(zeallot)) {
  y <- letters[1:4]
  c(Y, decode) %<-% onehot_with_decoder(y)
  Y
  decode(Y)
  identical(y, decode(Y))
  decode(Y[2,,drop = TRUE])
  decode(Y[2,,drop = FALSE])
  decode(Y[2:3,])

  rm(Y, decode)
}

# more peicemeal functions
Y <- onehot(y)
decode_onehot(Y)
#> [1] "a" "b" "c" "d"

# if you need to decode a matrix that lost colnames,
# make your own decoder that remembers classes
my_decode <- onehot_decoder(Y)
colnames(Y) <- NULL
my_decode(Y)
#> [1] "a" "b" "c" "d"
decode_onehot(Y)
#> [1] 1 2 3 4

# factor and numeric vectors also accepted
onehot(factor(letters[1:4]))
#>      a b c d
#> [1,] 1 0 0 0
#> [2,] 0 1 0 0
#> [3,] 0 0 1 0
#> [4,] 0 0 0 1
onehot(4:8)
#>      4 5 6 7 8
#> [1,] 1 0 0 0 0
#> [2,] 0 1 0 0 0
#> [3,] 0 0 1 0 0
#> [4,] 0 0 0 1 0
#> [5,] 0 0 0 0 1