Nothing
R/dataLoaders.R
In modello: Homemade Deep Learning Library
##' R6 class representing an array dataset
##'
##' It loads the data from an R array
##' @author Filippo Monari
##' @export
DataSet.Array = R6Class(
"DataSet.Array",
public = list(
##' @description
##' Initialisation method.
##'
##' @param X array
##' @param indexes subset of indexes along the sliceing dimension
##' @param yi subset of indexes to return as targets
##' @param si index of slicing dimension
initialize = function (X, indexes, yi, si) {
private$.X = X
private$.indexes0 = indexes
private$.yi = yi
private$.si = si
private$.data.dim = dim(X)[-si]
private$.slices = slice.index(X, si)
self$reset()
},
##' @description
##' Returns the length of the dataset
##' @examples
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' ds$length() # 20
length = function () {
if (is.null(private$.indexes0)) {
nrow(private$.X)
} else {
length(private$.indexes0)
}
},
##' @description
##' Returns TRUE if the dataset has still data
##' @examples
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' ds$has.next() # TRUE
has.next = function () {
length(private$.indexes) > 0
},
##' @description
##' Reset the dataset to its initial state
##' @examples
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' ds$..get.data..()
##' ds$..get.data..()
##' ds$reset()
##' ds$..get.data..()
reset = function () {
if (is.null(private$.indexes0)) {
private$.indexes = 1:self$length()
} else {
private$.indexes = private$.indexes0
}
},
##' @description
##' Auxiliary method not to be called directly.
##' It pops an index from the dataset according to the given method
##'
##' @param mode popping mode
..pop.index.. = function (mode="head") {
stopifnot(self$has.next())
if (mode == "head") {
i = 1
} else if (mode == "tail") {
i = length(private$.indexes)
} else if (mode == "random") {
i = sample(1:length(private$.indexes), 1)
} else {
stop("Unkwown 'mode'.")
}
index = private$.indexes[i]
private$.indexes = private$.indexes[-i]
return(index)
},
##' @description
##' Auxiliary method not to be called directly.
##'
##' @param mode popping mode
..get.data.. = function (mode="head") {
iii = which(private$.slices == self$..pop.index..(mode))
ANS = private$.X[iii]
dim(ANS) = private$.data.dim
return(ANS)
},
##' @description
##' Method to feed the bathces of data to 'numbers'.
##'
##' @param x number
##' @param mode popping mode
##' @examples
##' modello.init(100, 100, 100, 100)
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' x = number(rnorm(5))
##' print(x$v)
##' ds$feed(x)
##' print(x$v)
##' modello.close()
feed = function (x, mode="head") {
stopifnot(all(dim(x) == private$.data.dim))
x$set.v(self$..get.data..(mode))
}
),
private = list(
.X = NULL,
.indexes0 = NULL,
.indexes = c(),
.yi = NULL,
.si = NULL,
.data.dim = NULL,
.slices = NULL
)
)
##' R6 class representing a LaF csv data loader
##'
##' It uses the capability of the LaF package to load
##' sequentially batch of data from a csv file.
##' @author Filippo Monari
##' @export
DataSet.CSV = R6Class(
'DataSet.CSV',
public = list(
##' @field df stores the batch of data read from the csv file
df = NULL,
##' @description
##' Initialisation method.
##'
##' @param inf path to file or a LaF csv data model.
##' @param n number of rows to read ecah time from the csv file
##' @param yi indexes of the output variables
##' @param si index of slicing dimension after feature engineering
##' @param ... additional arguments for the
##' \code{LaF::detect_dm_csv} function
initialize = function (inf, n, yi, ...) {
if (is.character(inf)) {
private$.dm = detect_dm_csv(inf, ...)
} else if (islist(inf)) {
private$.dm = inf
} else {
stop("'inf' must be the path to the csv file or a LaF csv data model.")
}
private$.nrows = n
private$.yi = yi
},
##' @description
##' Opens the connection with the csv file.
##'
##' @param ... additional arguments for the
##' \code{LaF::laf_open} function
..connect.. = function (...) {
private$.conn = laf_open(private$.dm, ...)
},
##' @description
##' Return a batch of data of the wanted size.
##'
##' @param n size of the batch of data
##' @param ... additional arguments for the process fucntion
..get.data.. = function () {
if (is.null(self$df) || nrow(self$df) < private$.nrows) {
private$.load()
}
df = self$df[1:private$.nrows,]
self$df = self$df[-(1:private$.nrows),]
df
},
##' @description
##' Feed a number with the data from the CSV file
##' Returns X invisibly.
##'
##' @param data list composed by X (containing the features) and Y (containing the targets).
##' If NULL a the list is created. If provided the numbers in ti will be fed in place.
##' @param ... additional arguemnts to be passed to
feed = function (data=NULL, ...) {
DATA = as.matrix(self$..get.data..())
if (is.null(X)) {
data = list(X=number(DATA[,-yi], dx=FALSE), Y=number(DATA[,yi], dx=FALSE))
} else {
data$X$set.v(DATA[,-yi])
data$X$set.v(DATA[,yi])
}
invisible(data)
},
##' @description
##' Returns the data model.
dm = function () {
private$.dm
}
),
private = list(
.dm = NULL,
.conn = NULL,
.nrows = NULL,
.load = function () {
df = next_block(private$.conn, nrows=private$.nrows)
if (nrow(df) < private$.nrows) {
begin(private$.conn)
df = rbind(df, next_block(private$.conn, nrows=private$.nrows))
}
self$df = rbind(self$df, df)
}
)
)
##' R6 class representing a cluster LaF csv data loaders
##'
##' It load batch of data from a set of csv files and
##' binds them together.
##' @author Filippo Monari
##' @export
DataLoader.CSV = R6Class(
'DataLoader.CSV',
public = list(
##' @description
##' Initialisation method
##'
##' @param loaders a list containing the 'CSVLoader.LaF' objects.
initialize = function (loaders) {
private$.loaders = loaders
},
##' @description
##' Calls the connect methods of the loadres in order to
##' open the connections with the csv files
##'
##' @param ... additional argument for the \code{connect}
##' methods of each loader.
..connect.. = function (...) {
for (ld in private$.loaders) ld$connect(...)
},
##' @description
##' Returns the binded batch of data after applying the
##' \code{self$process} function.
##'
##' @param n number of rows to draw from each csv file
##' @param ... additional arguments for the \code{self$process} function
..get.data.. = function (n, ...) {
foreach(ld = private$.loaders, .combine=rbind) %dopar% ld$..get.data..()
},
##' @description
##' Feed a number with the data from the CSV file
##' Returns X invisibly.
##'
##' @param data list composed by X (containing the features) and Y (containing the targets).
##' If NULL a the list is created. If provided the numbers in ti will be fed in place.
##' @param ... additional arguemnts to be passed to
feed = function (data=NULL, ...) {
DATA = as.matrix(self$..get.data..())
if (is.null(X)) {
data = list(X=number(DATA[,-yi], dx=FALSE), Y=number(DATA[,yi], dx=FALSE))
} else {
data$X$set.v(DATA[,-yi])
data$X$set.v(DATA[,yi])
}
invisible(data)
},
##' @description
##' Returns a list with the loaders.
loaders = function () {
private$.loaders
}
),
private = list(
.loaders = NULL
)
)
##' R6 class representign a dataset of sequences to feed RNN
##'
##' It load the data from a R array
##' @author Filippo Monari
##' @export
DataSet.Seq = R6Class(
inherit = DataSet.Array,
public = list(
##' @description
##' Initialisation method
##'
##' @param X matrix/array
##' @param ls integer, sequence length
##' @param yi integer vector, subset of indices to return a targets
##' @param indexes integer vector, a subset of indexes
##' indicating the starting point fir the sequences
initialize = function (X, ls, yi, indexes=NULL) {
private$.X = X
private$.ls = ls
private$.yi = yi
private$.indexes0 = indexes
self$reset()
},
##' @description
##' Returns the length of the dataset
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' ds$length() # 20
length = function () {
if (is.null(private$.indexes0)) {
nrow(private$.X) - private$.ls
} else {
length(private$.indexes0)
}
},
##' @description
##' Feeds the data into an existing seqience or creating a new one
##'
##' @param x existing sequence, if NULL it creates a new one
##' @param mode chectacter indicating the way to build the sequences
##' @examples
##' modello.init(100, 100, 100, 100)
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' SEQ = ds$feed(NULL)
##' print(names(SEQ))
##' print(SEQ$x[[1]]$v)
##' print(SEQ$y$v)
##' ds$feed(SEQ)
##' print(SEQ$x[[1]]$v)
##' print(SEQ$y$v)
##' modello.close()
feed = function (x, mode="head") {
if (is.null(x)) {
ANS = self$..get.data..(mode)
x = lapply(ANS$x, function(x) number(as.matrix(x), dx=FALSE))
y = number(as.matrix(ANS$y), dx=FALSE)
return(list(x=x, y=y))
}
x$y$set.v(self$..feed..(x$x, mode))
return(invisible(x))
},
##' @description
##' Sets the data loader back of n previous states.
##' Only available for "head" and "tail" modes.
##'
##' @param n number of step to rewind
##' @param mode rewind mode. Possibly the same as the feed mode.
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' ds$..get.data..("head")
##' ds$..get.data..("head")
##' ds$rewind(2)
##' ds$..get.data..("head")
##' ds$..get.data..("head")
rewind = function (n, mode="head") {
stopifnot(length(private$.tape) >= n)
for (i in 1:n) {
if (mode == "head") {
private$.indexes = c(private$.tape[1], private$.indexes)
private$.tape = private$.tape[-1]
} else if (mode == "tail") {
private$.indexes = c(private$.indexes, private$.tape[1])
private$.tape = private$.tape[-1]
} else {
stop("Unkown mode. Rewind is only allowed for 'head' or 'tail'")
}
}
},
##' @description
##' Auxiliary method not to be called directly.
##' It pops an index from the dataset according to the given method
##'
##' @param mode popping mode
..pop.index.. = function (mode="head") {
stopifnot(self$has.next())
if (mode == "head") {
i = 1
private$.tape = c(private$.tape, private$.indexes[i])
} else if (mode == "tail") {
i = length(private$.indexes)
private$.tape = c(private$.tape, private$.indexes[i])
} else if (mode == "random") {
i = sample(1:length(private$.indexes), 1)
} else {
stop("Unkwown 'mode'.")
}
index = private$.indexes[i]
private$.indexes = private$.indexes[-i]
return(index)
},
##' @description
##' Auxiliary method. Not to be called directly.
##'
##' @param x existing sequence, if NULL it creates a new one
##' @param mode chectacter indicating the way to build the sequences
..feed.. = function (x, mode) {
stopifnot(length(x) == private$.ls)
n = private$.ls - 1
a = self$..pop.index..(mode)
b = a + n
for (i in a:b) {
x[[1]]$set.v(private$.X[i,])
x = x[-1]
}
return(private$.X[b+1,private$.yi])
},
##' @description
##' Auxiliary method. Not to be called directly.
##'
##' @param mode chectacter indicating the way to build the sequences
..get.data.. = function (mode) {
n = private$.ls - 1
a = self$..pop.index..(mode)
b = a + n
list(x=lapply(a:b, function(i)private$.X[i,]),
y=private$.X[b+1,private$.yi])
}
),
private = list(
.X = NULL,
.ls = NULL,
.yi = NULL,
.indexes0 = NULL,
.indexes = c(),
.tape = c()
)
)
##' R6 class representign a data loader for sequence datasets.
##'
##' It loads the data from a R array
##' @author Filippo Monari
##' @export
DataLoader.Seq = R6Class(
public = list(
##' @description
##' Initialisation method
##'
##' @param ds sequence datast
##' @param bz integer, batch size
initialize = function (ds, bz) {
private$.ds = ds
private$.bz = bz
self$reset()
},
##' @description
##' Returns the length of teh data loader object
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' dl = DataLoader.Seq$new(ds, 5)
##' ds$length()
##' dl$length()
length = function () {
floor(private$.ds$length() / private$.bz)
},
##' @description
##' Returns TRUE if the data loader has still batches available
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' dl = DataLoader.Seq$new(ds, 5)
##' ds$has.next() # TRUE
has.next = function () {
private$.batchi < self$length()
},
##' @description
##' Resets the data loader to its initial state
reset = function () {
private$.ds$reset()
private$.batchi = 0
},
##' @description
##' Feeds the data into a sequence of batches of sequences or it creates a new one
##'
##' @param SEQ batch of sequences. If NULL it creates a new one
##' @param mode feeding mode
##' @examples
##' modello.init(100, 100, 100, 100)
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' dl = DataLoader.Seq$new(ds, 5)
##' SEQ = dl$feed(NULL)
##' print(SEQ)
##' print(SEQ$X[[1]][[1]]$v)
##' dl$feed(SEQ)
##' print(SEQ$X[[1]][[1]]$v)
##' modello.close()
feed = function (SEQ, mode="head") {
stopifnot(self$has.next())
if (is.null(SEQ)) {
ANS = private$.feed0(mode)
} else {
stopifnot(length(SEQ$X) == private$.bz)
SEQ$Y$set.v(sapply(SEQ$X, function(x)private$.ds$..feed..(x, mode)))
ANS = SEQ
}
private$.batchi = private$.batchi + 1
invisible(ANS)
}
),
private = list(
.ds = NULL,
.bz = NULL,
.batchi = NULL,
.feed0 = function (mode="head") {
ANS = lapply(1:private$.bz, function(i) private$.ds$..get.data..(mode))
X = lapply(ANS, function (x) {
lapply(x$x, function (xx) {
number(as.matrix(xx), dx=FALSE)
})
})
Y = number(sapply(ANS, function(x)x$y), dx=FALSE)
list("X"=X, "Y"=Y)
}
)
)
Try the modello package in your browser
Any scripts or data that you put into this service are public.
modello documentation built on Feb. 2, 2021, 9:06 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
##' R6 class representing an array dataset
##'
##' It loads the data from an R array
##' @author Filippo Monari
##' @export
DataSet.Array = R6Class(
"DataSet.Array",
public = list(
##' @description
##' Initialisation method.
##'
##' @param X array
##' @param indexes subset of indexes along the sliceing dimension
##' @param yi subset of indexes to return as targets
##' @param si index of slicing dimension
initialize = function (X, indexes, yi, si) {
private$.X = X
private$.indexes0 = indexes
private$.yi = yi
private$.si = si
private$.data.dim = dim(X)[-si]
private$.slices = slice.index(X, si)
self$reset()
},
##' @description
##' Returns the length of the dataset
##' @examples
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' ds$length() # 20
length = function () {
if (is.null(private$.indexes0)) {
nrow(private$.X)
} else {
length(private$.indexes0)
}
},
##' @description
##' Returns TRUE if the dataset has still data
##' @examples
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' ds$has.next() # TRUE
has.next = function () {
length(private$.indexes) > 0
},
##' @description
##' Reset the dataset to its initial state
##' @examples
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' ds$..get.data..()
##' ds$..get.data..()
##' ds$reset()
##' ds$..get.data..()
reset = function () {
if (is.null(private$.indexes0)) {
private$.indexes = 1:self$length()
} else {
private$.indexes = private$.indexes0
}
},
##' @description
##' Auxiliary method not to be called directly.
##' It pops an index from the dataset according to the given method
##'
##' @param mode popping mode
..pop.index.. = function (mode="head") {
stopifnot(self$has.next())
if (mode == "head") {
i = 1
} else if (mode == "tail") {
i = length(private$.indexes)
} else if (mode == "random") {
i = sample(1:length(private$.indexes), 1)
} else {
stop("Unkwown 'mode'.")
}
index = private$.indexes[i]
private$.indexes = private$.indexes[-i]
return(index)
},
##' @description
##' Auxiliary method not to be called directly.
##'
##' @param mode popping mode
..get.data.. = function (mode="head") {
iii = which(private$.slices == self$..pop.index..(mode))
ANS = private$.X[iii]
dim(ANS) = private$.data.dim
return(ANS)
},
##' @description
##' Method to feed the bathces of data to 'numbers'.
##'
##' @param x number
##' @param mode popping mode
##' @examples
##' modello.init(100, 100, 100, 100)
##' X = matrix(rnorm(100), 20)
##' ds = DataSet.Array$new(X, 1:20, 1, 1)
##' x = number(rnorm(5))
##' print(x$v)
##' ds$feed(x)
##' print(x$v)
##' modello.close()
feed = function (x, mode="head") {
stopifnot(all(dim(x) == private$.data.dim))
x$set.v(self$..get.data..(mode))
}
),
private = list(
.X = NULL,
.indexes0 = NULL,
.indexes = c(),
.yi = NULL,
.si = NULL,
.data.dim = NULL,
.slices = NULL
)
)
##' R6 class representing a LaF csv data loader
##'
##' It uses the capability of the LaF package to load
##' sequentially batch of data from a csv file.
##' @author Filippo Monari
##' @export
DataSet.CSV = R6Class(
'DataSet.CSV',
public = list(
##' @field df stores the batch of data read from the csv file
df = NULL,
##' @description
##' Initialisation method.
##'
##' @param inf path to file or a LaF csv data model.
##' @param n number of rows to read ecah time from the csv file
##' @param yi indexes of the output variables
##' @param si index of slicing dimension after feature engineering
##' @param ... additional arguments for the
##' \code{LaF::detect_dm_csv} function
initialize = function (inf, n, yi, ...) {
if (is.character(inf)) {
private$.dm = detect_dm_csv(inf, ...)
} else if (islist(inf)) {
private$.dm = inf
} else {
stop("'inf' must be the path to the csv file or a LaF csv data model.")
}
private$.nrows = n
private$.yi = yi
},
##' @description
##' Opens the connection with the csv file.
##'
##' @param ... additional arguments for the
##' \code{LaF::laf_open} function
..connect.. = function (...) {
private$.conn = laf_open(private$.dm, ...)
},
##' @description
##' Return a batch of data of the wanted size.
##'
##' @param n size of the batch of data
##' @param ... additional arguments for the process fucntion
..get.data.. = function () {
if (is.null(self$df) || nrow(self$df) < private$.nrows) {
private$.load()
}
df = self$df[1:private$.nrows,]
self$df = self$df[-(1:private$.nrows),]
df
},
##' @description
##' Feed a number with the data from the CSV file
##' Returns X invisibly.
##'
##' @param data list composed by X (containing the features) and Y (containing the targets).
##' If NULL a the list is created. If provided the numbers in ti will be fed in place.
##' @param ... additional arguemnts to be passed to
feed = function (data=NULL, ...) {
DATA = as.matrix(self$..get.data..())
if (is.null(X)) {
data = list(X=number(DATA[,-yi], dx=FALSE), Y=number(DATA[,yi], dx=FALSE))
} else {
data$X$set.v(DATA[,-yi])
data$X$set.v(DATA[,yi])
}
invisible(data)
},
##' @description
##' Returns the data model.
dm = function () {
private$.dm
}
),
private = list(
.dm = NULL,
.conn = NULL,
.nrows = NULL,
.load = function () {
df = next_block(private$.conn, nrows=private$.nrows)
if (nrow(df) < private$.nrows) {
begin(private$.conn)
df = rbind(df, next_block(private$.conn, nrows=private$.nrows))
}
self$df = rbind(self$df, df)
}
)
)
##' R6 class representing a cluster LaF csv data loaders
##'
##' It load batch of data from a set of csv files and
##' binds them together.
##' @author Filippo Monari
##' @export
DataLoader.CSV = R6Class(
'DataLoader.CSV',
public = list(
##' @description
##' Initialisation method
##'
##' @param loaders a list containing the 'CSVLoader.LaF' objects.
initialize = function (loaders) {
private$.loaders = loaders
},
##' @description
##' Calls the connect methods of the loadres in order to
##' open the connections with the csv files
##'
##' @param ... additional argument for the \code{connect}
##' methods of each loader.
..connect.. = function (...) {
for (ld in private$.loaders) ld$connect(...)
},
##' @description
##' Returns the binded batch of data after applying the
##' \code{self$process} function.
##'
##' @param n number of rows to draw from each csv file
##' @param ... additional arguments for the \code{self$process} function
..get.data.. = function (n, ...) {
foreach(ld = private$.loaders, .combine=rbind) %dopar% ld$..get.data..()
},
##' @description
##' Feed a number with the data from the CSV file
##' Returns X invisibly.
##'
##' @param data list composed by X (containing the features) and Y (containing the targets).
##' If NULL a the list is created. If provided the numbers in ti will be fed in place.
##' @param ... additional arguemnts to be passed to
feed = function (data=NULL, ...) {
DATA = as.matrix(self$..get.data..())
if (is.null(X)) {
data = list(X=number(DATA[,-yi], dx=FALSE), Y=number(DATA[,yi], dx=FALSE))
} else {
data$X$set.v(DATA[,-yi])
data$X$set.v(DATA[,yi])
}
invisible(data)
},
##' @description
##' Returns a list with the loaders.
loaders = function () {
private$.loaders
}
),
private = list(
.loaders = NULL
)
)
##' R6 class representign a dataset of sequences to feed RNN
##'
##' It load the data from a R array
##' @author Filippo Monari
##' @export
DataSet.Seq = R6Class(
inherit = DataSet.Array,
public = list(
##' @description
##' Initialisation method
##'
##' @param X matrix/array
##' @param ls integer, sequence length
##' @param yi integer vector, subset of indices to return a targets
##' @param indexes integer vector, a subset of indexes
##' indicating the starting point fir the sequences
initialize = function (X, ls, yi, indexes=NULL) {
private$.X = X
private$.ls = ls
private$.yi = yi
private$.indexes0 = indexes
self$reset()
},
##' @description
##' Returns the length of the dataset
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' ds$length() # 20
length = function () {
if (is.null(private$.indexes0)) {
nrow(private$.X) - private$.ls
} else {
length(private$.indexes0)
}
},
##' @description
##' Feeds the data into an existing seqience or creating a new one
##'
##' @param x existing sequence, if NULL it creates a new one
##' @param mode chectacter indicating the way to build the sequences
##' @examples
##' modello.init(100, 100, 100, 100)
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' SEQ = ds$feed(NULL)
##' print(names(SEQ))
##' print(SEQ$x[[1]]$v)
##' print(SEQ$y$v)
##' ds$feed(SEQ)
##' print(SEQ$x[[1]]$v)
##' print(SEQ$y$v)
##' modello.close()
feed = function (x, mode="head") {
if (is.null(x)) {
ANS = self$..get.data..(mode)
x = lapply(ANS$x, function(x) number(as.matrix(x), dx=FALSE))
y = number(as.matrix(ANS$y), dx=FALSE)
return(list(x=x, y=y))
}
x$y$set.v(self$..feed..(x$x, mode))
return(invisible(x))
},
##' @description
##' Sets the data loader back of n previous states.
##' Only available for "head" and "tail" modes.
##'
##' @param n number of step to rewind
##' @param mode rewind mode. Possibly the same as the feed mode.
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' ds$..get.data..("head")
##' ds$..get.data..("head")
##' ds$rewind(2)
##' ds$..get.data..("head")
##' ds$..get.data..("head")
rewind = function (n, mode="head") {
stopifnot(length(private$.tape) >= n)
for (i in 1:n) {
if (mode == "head") {
private$.indexes = c(private$.tape[1], private$.indexes)
private$.tape = private$.tape[-1]
} else if (mode == "tail") {
private$.indexes = c(private$.indexes, private$.tape[1])
private$.tape = private$.tape[-1]
} else {
stop("Unkown mode. Rewind is only allowed for 'head' or 'tail'")
}
}
},
##' @description
##' Auxiliary method not to be called directly.
##' It pops an index from the dataset according to the given method
##'
##' @param mode popping mode
..pop.index.. = function (mode="head") {
stopifnot(self$has.next())
if (mode == "head") {
i = 1
private$.tape = c(private$.tape, private$.indexes[i])
} else if (mode == "tail") {
i = length(private$.indexes)
private$.tape = c(private$.tape, private$.indexes[i])
} else if (mode == "random") {
i = sample(1:length(private$.indexes), 1)
} else {
stop("Unkwown 'mode'.")
}
index = private$.indexes[i]
private$.indexes = private$.indexes[-i]
return(index)
},
##' @description
##' Auxiliary method. Not to be called directly.
##'
##' @param x existing sequence, if NULL it creates a new one
##' @param mode chectacter indicating the way to build the sequences
..feed.. = function (x, mode) {
stopifnot(length(x) == private$.ls)
n = private$.ls - 1
a = self$..pop.index..(mode)
b = a + n
for (i in a:b) {
x[[1]]$set.v(private$.X[i,])
x = x[-1]
}
return(private$.X[b+1,private$.yi])
},
##' @description
##' Auxiliary method. Not to be called directly.
##'
##' @param mode chectacter indicating the way to build the sequences
..get.data.. = function (mode) {
n = private$.ls - 1
a = self$..pop.index..(mode)
b = a + n
list(x=lapply(a:b, function(i)private$.X[i,]),
y=private$.X[b+1,private$.yi])
}
),
private = list(
.X = NULL,
.ls = NULL,
.yi = NULL,
.indexes0 = NULL,
.indexes = c(),
.tape = c()
)
)
##' R6 class representign a data loader for sequence datasets.
##'
##' It loads the data from a R array
##' @author Filippo Monari
##' @export
DataLoader.Seq = R6Class(
public = list(
##' @description
##' Initialisation method
##'
##' @param ds sequence datast
##' @param bz integer, batch size
initialize = function (ds, bz) {
private$.ds = ds
private$.bz = bz
self$reset()
},
##' @description
##' Returns the length of teh data loader object
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' dl = DataLoader.Seq$new(ds, 5)
##' ds$length()
##' dl$length()
length = function () {
floor(private$.ds$length() / private$.bz)
},
##' @description
##' Returns TRUE if the data loader has still batches available
##' @examples
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' dl = DataLoader.Seq$new(ds, 5)
##' ds$has.next() # TRUE
has.next = function () {
private$.batchi < self$length()
},
##' @description
##' Resets the data loader to its initial state
reset = function () {
private$.ds$reset()
private$.batchi = 0
},
##' @description
##' Feeds the data into a sequence of batches of sequences or it creates a new one
##'
##' @param SEQ batch of sequences. If NULL it creates a new one
##' @param mode feeding mode
##' @examples
##' modello.init(100, 100, 100, 100)
##' ds = DataSet.Seq$new(matrix(rnorm(100), 20), 5, 1)
##' dl = DataLoader.Seq$new(ds, 5)
##' SEQ = dl$feed(NULL)
##' print(SEQ)
##' print(SEQ$X[[1]][[1]]$v)
##' dl$feed(SEQ)
##' print(SEQ$X[[1]][[1]]$v)
##' modello.close()
feed = function (SEQ, mode="head") {
stopifnot(self$has.next())
if (is.null(SEQ)) {
ANS = private$.feed0(mode)
} else {
stopifnot(length(SEQ$X) == private$.bz)
SEQ$Y$set.v(sapply(SEQ$X, function(x)private$.ds$..feed..(x, mode)))
ANS = SEQ
}
private$.batchi = private$.batchi + 1
invisible(ANS)
}
),
private = list(
.ds = NULL,
.bz = NULL,
.batchi = NULL,
.feed0 = function (mode="head") {
ANS = lapply(1:private$.bz, function(i) private$.ds$..get.data..(mode))
X = lapply(ANS, function (x) {
lapply(x$x, function (xx) {
number(as.matrix(xx), dx=FALSE)
})
})
Y = number(sapply(ANS, function(x)x$y), dx=FALSE)
list("X"=X, "Y"=Y)
}
)
)
Try the modello package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.