Skip to contents

Some models have several several vectors of parameters, possibly of different lengths, such that in some circumstances they are thought of as lists, in others as matrices after suitable padding with zeroes. Class "raggedCoef" represents such ragged lists. In package "MixAR" it is used to hold the autoregressive coefficients of MixAR models.

Usage

raggedCoef(p, value = NA_real_)

Arguments

p

orders, vector of integers.

value

typically, a list, but see Details.

Details

Class "raggedCoef" is for objects that can be considered as both, lists and matrices. The elements of the list are vectors, possibly of different lengths. When the object is viewed as a matrix, each element of the list (suitably padded with zeroes or NAs) represents a row of a matrix.

The recommended way to create objects from class "raggedCoef" is with the function raggedCoef.

If value is a "raggedCoef" object it is returned. If value is a list, it is converted to "raggedCoef" using new(). If argument p is missing, it is inferred from the lengths of the elements of the list. If argument p is not missing, a consistency check is made to ensure that the order of the object is as specified by p.

Otherwise, if value is of length one, it is replicated to form a ragged list with i-th element a vector of length p[i]. Although not checked, the intention here is that value is from some atomic class. The default for value is NA_real_ to give a convenient way to create a ragged list.

Finally, if none of the above applies, value is effectively assumed to be a vector of length sum(p), although other cases are admissible (but I don't remember if this was intended). In this case, value is reshaped into a ragged list to match p. This is convenient when, for example, the elements of a ragged array are obtained from an optimisation routine which expects plain vector.

Objects from the Class

Below we describe the "initialize" method that underlies new("raggedCoef", ...). The recommended way to create "raggedCoef" objects is with the function raggedCoef, see section Details.

Objects can also be created by calls of the form new("raggedCoef", v), where v is a list whose elements are numeric vectors, or new("raggedCoef", v1, v2, ...), where v1, v2, ... are numeric vectors. The two forms are equivalent if v = list(v1, v2, ...).

The elements of the list v may be named. Similarly, named arguments can be used in the second form, say new("raggedCoef", name1 = v1, name2 = v2, ...). In both cases the names are preserved in the internal representaion, but not used.

If the arguments are not as specified above the result should be considered undefined. Currently, if there are other arguments after the list v, they are ignored with a warning. If the first argument is not a list then all arguments must be numeric and an error is raised if this is not the case. For completeness, we mention that exactly two arguments named a, and p are also accepted by new(), eg new("raggedCoef", p = c(1, 2), a = list(3, 4:5)), but these are assigned to the slots without any checking. so it is most flexible (and recommended) to use raggedCoef() instead.

Slots

a:

Object of class "list" containing the values.

p:

Object of class "numeric" containing the lengths of the components of a.

Methods

Indexing with "[" treats a raggedCoef object as a matrix, while "[[" treats the object as list (it works on slot a).

Note that there is a difference between x[2,] (or the equivalent x[2]) and x[[2]]---the former gives a vector of length max(p), so potentially padded with zeroes, while the latter gives the component with its ``natural'' length.

The replacement variants of "[" and "[[" do not change the structure of the object and issue errors if the replacement value would result in that. In situations where the checks are deemed redundant, direct assignments to the corresponding slots may be used.

[

signature(x = "raggedCoef", i = "missing", j = "missing", drop = "ANY"):

[

signature(x = "raggedCoef", i = "missing", j = "numeric", drop = "ANY"):

[

signature(x = "raggedCoef", i = "numeric", j = "missing", drop = "ANY"):

[

signature(x = "raggedCoef", i = "numeric", j = "numeric", drop = "ANY"): Indexing with "[" treats a raggedCoef object as a matrix with one row for each component and number of columns equal to max(p). However, x[2] is equivalent to x[2,] which is different from the treatment of matrix objects in base R.

[[

signature(x = "raggedCoef", i = "ANY", j = "missing"):

[[

signature(x = "raggedCoef", i = "ANY", j = "ANY"): "[[" extracts the corresponding element of slot a.

[[<-

signature(x = "raggedCoef", i = "ANY", j = "ANY", value = "numeric"): Replace the j-th element of i-th row with value. All arguments must be scalars.

[[<-

signature(x = "raggedCoef", i = "ANY", j = "missing", value = "numeric"):

[<-

signature(x = "raggedCoef", i = "ANY", j = "missing", value = "numeric"): Replace the i-th row with value. Argument i must be a scalar while the length of value must be the same as that of x@a[[i]]. The methods for "[" and "[[" with this signature coinside.

[<-

signature(x = "raggedCoef", i = "ANY", j = "missing", value = "list"): The elements of value must have the same lengths as the elements they are replacing.

[<-

signature(x = "raggedCoef", i = "ANY", j = "missing", value = "matrix"): This is essentially the reverse od the corresponding non-replacement operator. value must have at least as many columns as the longest element of x that is replaced.

[<-

signature(x = "raggedCoef", i = "ANY", j = "ANY", value = "numeric"): ...

[<-

signature(x = "raggedCoef", i = "missing", j = "missing", value = "list"): ...

[<-

signature(x = "raggedCoef", i = "missing", j = "missing", value = "matrix"): ...

[<-

signature(x = "raggedCoef", i = "missing", j = "missing", value = "numeric"): ...

initialize

signature(.Object = "raggedCoef"): Creates objects of class raggedCoef. This method is used internally by new(). Users should use new() for creation of objects from this class, see the examples.

show

signature(object = "raggedCoef"): ...

mixFilter

signature(x = "numeric", coef = "raggedCoef", index = "numeric"): Apply a mixture filter to a time series.

row_lengths

signature(x = "raggedCoef"): Gives x@p, which is the same as lengths(x@a).

length

signature(x = "raggedCoef"): Gives the total number of coefficients (sum(x@p)).

anyNA

signature(x = "raggedCoef"): Are there NA's in x@a?

dim

signature(x = "raggedCoef"): The dimension of the object, when viewed as a matrix. The presence of this method also ensures that nrow() and related functions give the expected result.

Author

Georgi N. Boshnakov

Note

Slot p is redundant but convenient.

See also

class "MixARGaussian"

Examples

ragged1 <- list(1, 2:3, 4:6)
ragged2 <- list(a = 1, b = 2:3, c = 4:6)

raggedCoef(1:3)          # only order given, fill with NA's
#> An object of class "raggedCoef"
#> Number of rows: 3 
#> Components' lengths: 1 2 3 
#> 
#>             co_1 co_2 co_3
#> Component_1  NA           
#> Component_2  NA   NA      
#> Component_3  NA   NA   NA 
#> 
raggedCoef(1:3, 0)       # fill with a number (zero in this case)
#> An object of class "raggedCoef"
#> Number of rows: 3 
#> Components' lengths: 1 2 3 
#> 
#>             co_1 co_2 co_3
#> Component_1   0           
#> Component_2   0    0      
#> Component_3   0    0    0 
#> 

## init with a list
raggedCoef(ragged1)
#> An object of class "raggedCoef"
#> Number of rows: 3 
#> Components' lengths: 1 2 3 
#> 
#>             co_1 co_2 co_3
#> Component_1   1           
#> Component_2   2    3      
#> Component_3   4    5    6 
#> 
raggedCoef(value = ragged1) 
#> An object of class "raggedCoef"
#> Number of rows: 3 
#> Components' lengths: 1 2 3 
#> 
#>             co_1 co_2 co_3
#> Component_1   1           
#> Component_2   2    3      
#> Component_3   4    5    6 
#> 

## error, since the shape of ragged1 is not c(2, 2, 3):
## raggedCoef(c(2, 2, 3), value = ragged1)

## init with a flattened list
raggedCoef(p = 1:3, value = 1:6)
#> An object of class "raggedCoef"
#> Number of rows: 3 
#> Components' lengths: 1 2 3 
#> 
#>             co_1 co_2 co_3
#> Component_1   1           
#> Component_2   2    3      
#> Component_3   4    5    6 
#> 

## specify each component separately
ragA <- new("raggedCoef", 1, 2:3, 4:6)
ragB <- new("raggedCoef", list(1, 2:3, 4:6))  # same
identical(ragA, ragB) #TRUE
#> [1] TRUE

## extract as a matrix
ragA[]
#>      [,1] [,2] [,3]
#> [1,]    1    0    0
#> [2,]    2    3    0
#> [3,]    4    5    6

## extract the 2nd component
ragA[2]      # c(2, 3, 0) ("[" pads with 0's)
#> [1] 2 3 0
ragA[[2]]    # c(2, 3)    ("[[" does not pad)
#> [1] 2 3

## get the 2nd and 3rd components as a matrix
ragA[2:3, ]    # "[" treats object (almost) as matrix
#>      [,1] [,2] [,3]
#> [1,]    2    3    0
#> [2,]    4    5    6
ragA[2:3]      # same (though not as for "matrix")
#>      [,1] [,2] [,3]
#> [1,]    2    3    0
#> [2,]    4    5    6

## names are kept in the list but currently not used 
ragC  <- new("raggedCoef", list(a = 1, b = 2:3, c = 4:6))
ragC1 <- new("raggedCoef", a = 1, b = 2:3, c = 4:6)
identical(ragC, ragC1) # TRUE
#> [1] TRUE
names(ragC@a) # [1] "a" "b" "c"
#> [1] "a" "b" "c"

length(ragA)
#> [1] 6
dim(ragA)
#> [1] 3 3
c(nrow(ragA), ncol(ragA))
#> [1] 3 3
c(NROW(ragA), NCOL(ragA))
#> [1] 3 3