R - Create cut-like intervals with non-empty intersection

Question

I have a dataframe df with a column named x1 with values between -5 and +5. I am trying to assign for each row of df an interval regarding the values of x1 . The function cut allow me do to so :

cut(df$x1,c(-5,-4,-3,-2,-1,0,1,2,3,4,5))

and I can then split df into 10 data.frames using by . Unfortunately what I would like is to assign intervals like -5 to -3.95, -4.05 to -2.95, -3.05 to -1.95 and so on, meaning that :

• 4.06 will be in the interval 3.95-5.05
• 4.05 will be in the interval 3.95-5.05 and 2.95-4.05
• 4.04 will be in the interval 3.95-5.05 and 2.95-4.05
• 3.94 will be in the interval 2.95-4.05

which means that after using by I will have 10 dataframes with a few elements in 2 of those dataframes.

The next part of my question would concern the values near 0 : the intervals should not contain negative and positive values, so the intervals would be like

• -5 to -3.95
• -4.05 to -2.95
• ...
• -2.05 to -0.95
• -1.05 to 0 AND NOT -1.05 to 0.05
• 0 to 1.05 AND NOT -0.05 to 1.05
• 0.95 to 2.05
• ...

Is there a way to achieve that in R ?

EDIT : df

df looks like this :

other_var  ...   x1  ... another_var ...
    100    ... 4     ...   18     ...
    12.3   ... 3.84  ...   -6.2   ...
    1.4    ... 4.78  ...    4.78  ...
    -2     ... -2.51 ...    7.1   ...
    -3.2   ... 0.57  ...   -1     ...


dput(df1)

structure(list(x0 = c(0.702166747375488, 0.205532096598193,     0.0704982518296982, 
-0.159150628995597, -0.162625494967927, -0.331660025490033, -0.099135847436449, 
-0.137985446193678, -0.179304942878067, 0.0554309512268647), 
x1 = c(-0.561621170364712, -0.762747775318984, 1.63791710226613, 
-0.861210697757564, -1.05393723031543, 0.809872536189693, 
2.85973319518198, 0.211750306033687, 1.18360826959114, -0.358159130198865
), x2 = c(-0.304711385106637, 0.365667729645747, -0.406328268107825, 
-0.315315872233279, -0.477546612710489, 0.251158976293131, 
-1.1263800774781, 0.229002212764429, -0.00413111289214729, 
-0.252467704090853)), .Names = c("x0", "x1", "x2"), row.names = c(NA, 
10L), class = "data.frame")

Answer 1

I could not see a solution with creating intervals with cut that did not lead to multiple columns, so I approached it from another angle: iterate over all cutpoints and return the subset for that min and max.

intervals <- data.frame(min=c(-5,-4.05+0:3,0,0.95+0:3))
intervals$max <- rev(intervals$min)*-1
intervals$name <- with(intervals, sprintf("[%.2f;%.2f)",min,max))
res <- lapply(split(intervals,intervals$name), function(x){
  return(df1[df1$x1> x$min & df1$x1 <= x$max,])
})

> head(res)
$`[-1.05;-0.00)`
            x0         x1         x2
1   0.70216675 -0.5616212 -0.3047114
2   0.20553210 -0.7627478  0.3656677
4  -0.15915063 -0.8612107 -0.3153159
10  0.05543095 -0.3581591 -0.2524677

$`[-2.05;-0.95)`
          x0        x1         x2
5 -0.1626255 -1.053937 -0.4775466

$`[-3.05;-1.95)`
[1] x0 x1 x2
<0 rows> (or 0-length row.names)

$`[-4.05;-2.95)`
[1] x0 x1 x2
<0 rows> (or 0-length row.names)

$`[-5.00;-3.95)`
[1] x0 x1 x2
<0 rows> (or 0-length row.names)

$`[0.00;1.05)`
          x0        x1        x2
6 -0.3316600 0.8098725 0.2511590
8 -0.1379854 0.2117503 0.2290022

Answer 2

Here's a solution that uses foverlaps(...) in the data.table package. Unfortunately. you need the most recent developmental version for this to work. Uses the intervals data.frame from the other answer.

##install.packages("devtools")
# library(devtools)
# install_github("Rdatatable/data.table", build_vignettes = FALSE)

library(data.table)
y    <- with(df1,data.table(row=1:nrow(df1),lo=x1, hi=x1, key=c("lo","hi")))
cuts <- foverlaps(setDT(intervals),y, by.x=c("min","max"))[,list(row,name)]
lapply(split(cuts, cuts$name),function(s)df1[sort(s$row),]) 
# $`[-1.05;-0.00)`
#            x0         x1         x2
# 1   0.70216675 -0.5616212 -0.3047114
# 2   0.20553210 -0.7627478  0.3656677
# 4  -0.15915063 -0.8612107 -0.3153159
# 10  0.05543095 -0.3581591 -0.2524677
#
# $`[-2.05;-0.95)`
#           x0        x1         x2
# 5 -0.1626255 -1.053937 -0.4775466
#
# $`[-3.05;-1.95)`
# [1] x0 x1 x2
# <0 rows> (or 0-length row.names)
#...

foverlaps(x,y,...) does an "overlap join", that is, it finds all the records in y which which have overlaps in x . Overlaps are defined as values in a range between to columns in y (say, a and b), which overlap the corresponding range in two columns in x (say c and d). In this case we use, for x , the intervals data.frame (converted to a data.table), and for y , a data.table formed with the lo and hi columns both = df$x1 .