Find me on github
A class inheriting from data.frame featuring drop=FALSE as default, an 'end' keyword, 0 index for 'all' and a 'desc' field.
Despite very powerful, R language is mostly intended for interactive use. So in x[i,j] it is not straight to set such a value for i or j to emulate x[,j] or x[i,].
Using TRUE (which will get recycled) is a trick to obtain this.
T=TRUE
x=matrix(1:12, ncol=3)
identical(x[,3], x[T,3])## [1] TRUE
identical(x[,3], `[`(x,T,3))## [1] TRUE
z=y=x
x[,3] =0
y[T,3]=0
identical(x,y)## [1] TRUE
z=`[<-`(z,T,3,0)
identical(x,z)## [1] TRUE
This works also for a data.frame.
x=data.frame(matrix(1:12, ncol=3))
identical(x[,3], x[T,3])## [1] TRUE
identical(x[,3], `[`(x,T,3))## [1] TRUE
z=y=x
x[,3] =0
y[T,3]=0
identical(x,y)## [1] TRUE
z=`[<-`(z,T,3,0)
identical(x,z)## [1] TRUE
Also the default drop=TRUE feature means breaking some programmatic code, like simple nrow, ncol, which will fail if the dimensions are dropped.
All these features are blessed time savers in interactive code, but oblige to many if/else check, prone to errors, in programmatic use.
I developed a new data.frame S4 class, frame, inheriting from data.frame but exploiting the 0, unused in subsetting, to select all rows or columns, so x[0,j] and x[i,0] as x[,j] and x[i,].
Also drop=FALSE is used as the default value in place of drop=TRUE.
Besides it uses a Matlab like end operator to get the last rows or columns item, e.g.: x[3:end, 2:end].
A desc field can be added to annotate any data description.
This class inherits from the data.frame class: if a function works with a data.frame it is supposed to work with 'frame'. If your code requires a formal data.frame, you can obtain it from x frame with:
dfm(x)
This is a shortcut for as.data.frame which in turn will call the specialised function as.data.frame.frame.
## Create a frame, from scratch (with column names)
frame(x, col.names=letters[1:3], desc="Hello frame")## Frame 4x3
## : Hello frame
## a b c
## 1 1 5 0
## 2 2 6 0
## 3 3 7 0
## 4 4 8 0
frame(1:4, 5:8, 9:12, desc="Hello frame")## Frame 4x3: Hello frame
## 1:4 5:8 9:12
## 1 1 5 9
## 2 2 6 10
## 3 3 7 11
## 4 4 8 12
## or use a data.frame
x=data.frame(1:4, 5:8, 9:12)
frame(x, desc="Hello frame")## Frame 4x3: Hello frame
## X1.4 X5.8 X9.12
## 1 1 5 9
## 2 2 6 10
## 3 3 7 11
## 4 4 8 12
Note the option for frame only col.names. If you don't use it, data.frame by default checks column names, frame does not, as it gives a somewhat unpleasant view. If you need it, use frame(1:4, 5:8, 9:12, desc="Hello frame", check.names = TRUE).
Similarly a very annoying feature of data.frame is the automatic conversion of strings to factors. Here instead:
x=frame(a=1:26, b=letters)
class(x$b) ## [1] "character"
## ... while
x=data.frame(a=1:26, b=letters)
class(x$b) # unless you modified general R options## [1] "factor"
You migh find convenient start from matrices:
frame(matrix(1:90, ncol=3), desc='Hello frame')## Frame 30x3
## : Hello frame
## 1 2 3
## 1 1 31 61
## 2 2 32 62
## 3 3 33 63
## 4 4 34 64
## 5 5 35 65
## 6 6 36 66
## 7 7 37 67
## 8 8 38 68
## 9 9 39 69
## 10 10 40 70
## 11 11 41 71
## 12 12 42 72
## 13 13 43 73
## 14 14 44 74
## 15 15 45 75
## 16 16 46 76
## 17 17 47 77
## 18 18 48 78
## 19 19 49 79
## 20 20 50 80
## ...
as.frame(matrix(1:90, ncol=3), desc='Hello frame')## Frame 30x3: Hello frame
## V1 V2 V3
## 1 1 31 61
## 2 2 32 62
## 3 3 33 63
## 4 4 34 64
## 5 5 35 65
## 6 6 36 66
## 7 7 37 67
## 8 8 38 68
## 9 9 39 69
## 10 10 40 70
## 11 11 41 71
## 12 12 42 72
## 13 13 43 73
## 14 14 44 74
## 15 15 45 75
## 16 16 46 76
## 17 17 47 77
## 18 18 48 78
## 19 19 49 79
## 20 20 50 80
## ...
## use "col.names" argument, if you don't like the automatic naming Note the ... signalling that only the first 20 rows were printed.
Nothing is more boring than printing an object just to discover you get a useless time consuming gigantic print.
Let us see the standard subsetting.
(fr=as.frame(matrix(1:12, ncol=3)))## Frame 4x3
##
## V1 V2 V3
## 1 1 5 9
## 2 2 6 10
## 3 3 7 11
## 4 4 8 12
## Standard subsetting
fr[c(1,3),2:3]## Frame 2x2
##
## V2 V3
## 1 5 9
## 3 7 11
fr[2,]## Frame 1x3
##
## V1 V2 V3
## 2 2 6 10
fr[2,c('V1', 'V3')]## Frame 1x2
##
## V1 V3
## 2 2 10
fr[c(T,F,T,T),T]## Frame 3x3
##
## V1 V2 V3
## 1 1 5 9
## 3 3 7 11
## 4 4 8 12
# fr[2:3] # error not supported New features: subsetting without dropping dimensions by default!
## Non-standard behaviour:
## drop is false by default
fr[1,1]
fr[1,1, drop=T] #Traditional behaviour
fr[ ,1, drop=T] #Traditional behaviour
New features: `end` operator.
Inside brackets `end` is a keyword and it means the last row or column according to its position.
## Error: <text>:8:5: unexpected symbol
## 7:
## 8: New features
## ^
(fr=as.frame(matrix(1:12, ncol=3)))## Frame 4x3
##
## V1 V2 V3
## 1 1 5 9
## 2 2 6 10
## 3 3 7 11
## 4 4 8 12
fr[end,end] # 'end' is not quoted inside brackets!## Frame 1x1
##
## V3
## 4 12
## fr[end] # error not supported
## fr[2:end] # error not supported
fr[2:end,end:1]## Frame 3x3
##
## V3 V2 V1
## 2 10 6 2
## 3 11 7 3
## 4 12 8 4
## 'end' is always a keyword inside brackets
end=100
fr[, end]## Frame 4x1
##
## V3
## 1 9
## 2 10
## 3 11
## 4 12
Let us examine subsetting with zero index. As an index zero means all (like TRUE).
## Zero means all
fr[0,1]## Frame 4x1
##
## V1
## 1 1
## 2 2
## 3 3
## 4 4
fr[1,0]## Frame 1x3
##
## V1 V2 V3
## 1 1 5 9
fr[0,0] # whole frame## Frame 4x3
##
## V1 V2 V3
## 1 1 5 9
## 2 2 6 10
## 3 3 7 11
## 4 4 8 12
Anyway sequences starting with zero are kept with the old meaning.
## Traditional zero behaviour
## For a DF [0:3,0:2] is like [1:3,1:2], and for 'frame' too
fr[0:3,0:2] # Your old code won't break## Frame 3x2
##
## V1 V2
## 1 1 5
## 2 2 6
## 3 3 7
Consider this code snippet
size=function(x)
if(nrow(x)>100) print("I am a long series") else
print("I am a short series")It will break when a data.frame loses a dimension, due to the default drop=TRUE.
Because nrow does not apply to atomic objects.
df=as.data.frame(matrix(1:12, ncol=3))
fr=frame(df)
## All columns
size(df)## [1] "I am a short series"
## Take only the first col (we lose a dimension)
# size(df[,1])
# Error: argument is of length zero
## Take only the first col (now we don't lose original dimension)
size(fr[,1])## [1] "I am a short series"
Assume that fr represents real observations.
fr=as.frame(matrix(rnorm(16), ncol=4))Calculate the covariance matrix of some or all the fr columns. Generate a random frame like fr, calculate the same covariance matrix and obtain the differences.
dcov=function(fr,i,j){
v1=fr[i,j]
m=matrix(rnorm(nrow(fr) * ncol(fr)), ncol=ncol(fr))
v2=as.frame(m)[i,j]
var(v1)-var(v2)
}Get delta-cov for all columns in fr:
dcov(fr, 0, 0)## V1 V2 V3 V4
## V1 -1.20717271 0.33114372 -0.80376393 -0.04201971
## V2 0.33114372 0.24550832 -0.06600963 -0.19319950
## V3 -0.80376393 -0.06600963 0.12442672 0.73472394
## V4 -0.04201971 -0.19319950 0.73472394 -0.19146884
Only columns starting from second:
dcov(fr, 0, '2:end') #Quoting necessary!!## V2 V3 V4
## V2 -2.7786743 -0.8103209 -0.2197724
## V3 -0.8103209 1.0137194 1.4579046
## V4 -0.2197724 1.4579046 -2.8973433
Note: Without quoting, if you have a variable end=3, you are sending 2:3.
All columns, but remove last row:
dcov(fr, '-end', 0) #Quoting necessary!!## V1 V2 V3 V4
## V1 0.1870858 -0.2862340 0.3589954 0.2551383
## V2 -0.2862340 -2.4014519 -0.6133402 0.7794532
## V3 0.3589954 -0.6133402 -0.1734055 0.1492068
## V4 0.2551383 0.7794532 0.1492068 -0.2049813
Inside brackets you do not quote 'end' to use it as a keyword for last element.
fr=as.frame(matrix(1:16, nrow=4))
end=100
fr[2:end,] #'end' is the last row (4 not 100)## Frame 3x4
##
## V1 V2 V3 V4
## 2 2 6 10 14
## 3 3 7 11 15
## 4 4 8 12 16
Outside brackets, that is in a programmatic environment when using 'end' keyword indirectly in a variable assignment, quote end!
end=1
a=end
b='end'
## Without quoting, variable value 'end' is 1
fr[1:2,a]## Frame 2x1
##
## V1
## 1 1
## 2 2
## By quoting, variable value 'end' is a keyword
fr[1:2,b]## Frame 2x1
##
## V4
## 1 13
## 2 14
If the assignment is an expression contains 'end' the same rule applies to get it as a keyword: quote too.
a='3:end'
b=0
fr[a,b]## Frame 2x4
##
## V1 V2 V3 V4
## 3 3 7 11 15
## 4 4 8 12 16
In assignments, without quoting 'end', it will get the current value of end, if any and valid, or raise an error. Inside brackets it will be passed unevaluated and after converted to a keyword. Quoting in brackets means querying for the row/column named "end".
Bracket alternatives. These rules keep for bracket alternatives `[`(x,i,j) and `[<-`(x,i,j).
`[`(fr,end,end)## Frame 1x1
##
## V4
## 4 16
a='3:end'
`[`(fr,a,a)## Frame 2x2
##
## V3 V4
## 3 11 15
## 4 12 16
fr=as.frame(matrix(1:16, nrow=4))
names(fr)[3]='end'
fr[1,]## Frame 1x4
##
## V1 V2 end V4
## 1 1 5 9 13
Inside brackets nothing special happens. As a keyword 'end' is unquoted, while as a name:
fr[, c('V1','end')]## Frame 4x2
##
## V1 end
## 1 1 9
## 2 2 10
## 3 3 11
## 4 4 12
For assignments to a variable, outside brackets, quote twice.
a="'end'"
b="c('V1', 'end')"
fr[1:2,a]## Frame 2x1
##
## end
## 1 9
## 2 10
fr[1:2,b]## Frame 2x2
##
## V1 end
## 1 1 9
## 2 2 10
Obviously you can quote twice also with sQuote, dQuote or escaping inner quotation marks with "\".