xts split by week function specify first day of week as Sunday instead of default of Monday

Question

Applying the split function to a xts object by weeks groups rows into weekly chunks. The default days in the group are Monday to Sunday. What do I do if I want the days in the group to be from Sunday to Saturday?

library(xts)
idx <- as.Date("2018-3-1") + 0:14
v <- 1:15
x <- xts(v, idx)
group <- split(x, f = 'weeks')
group

Output:
[[1]]
           [,1]
2018-03-01    1  # Thursday
2018-03-02    2  # Friday
2018-03-03    3  # Saturday
2018-03-04    4  # Sunday

[[2]]
           [,1]
2018-03-05    5  # Monday
2018-03-06    6  # Tuesday
2018-03-07    7  # Wednesday
2018-03-08    8  # Thursday
2018-03-09    9  # Friday
2018-03-10   10  # Saturday
2018-03-11   11  # Sunday

[[3]]
           [,1]
2018-03-12   12  # Monday
2018-03-13   13  # Tuesday
2018-03-14   14  # Wednesday
2018-03-15   15  # Thursday

Desired Output:
[[1]]
           [,1]
2018-03-01    1  # Thursday
2018-03-02    2  # Friday
2018-03-03    3  # Saturday

[[2]]
           [,1]
2018-03-04    4  # Sunday
2018-03-05    5  # Monday
2018-03-06    6  # Tuesday
2018-03-07    7  # Wednesday
2018-03-08    8  # Thursday
2018-03-09    9  # Friday
2018-03-10   10  # Saturday

[[3]]
           [,1]
2018-03-11   11  # Sunday
2018-03-12   12  # Monday
2018-03-13   13  # Tuesday
2018-03-14   14  # Wednesday
2018-03-15   15  # Thursday

Answer 1

I split by weeks on Sundays rather than Mondays frequently, because I work with FX data (with markets opening on Sunday afternoon New York EST). Here is an efficient solution, split_FXweeks, using the "xts way" of splitting time series data. This approach is quite fast when you're working with high density tick data over long periods of time.

Credit for this trick is due to trick 1 in the below link: http://darrendev.blogspot.com.au/2012/08/small-rxts-code-snippets-and-tips.html

Added a benchmark comparing to other suggested approaches as a baseline.

idx <- as.Date("2018-3-1") + 0:14
v <- 1:15
x <- xts(v, idx)


split_FXweeks <- function(x) {
  ep <- .Call("endpoints", .index(x) + 4L * 86400L, 604800L, 
              1, TRUE, PACKAGE = "xts")
  sp <- (ep + 1)[-length(ep)]
  ep <- ep[-1]
  lapply(1:length(ep), function(X) x[sp[X]:ep[X]])
}


split1 <- function(idx, x) {
  week_num <- format(idx, "%U")
  group <- unname(split(x, f = week_num))
  group
}

library(microbenchmark)
microbenchmark(
  y <- split_FXweeks(x),
  z <- split1(idx, x))
# Unit: microseconds
# expr     min      lq      mean   median       uq     max neval
# y <- split_FXweeks(x)  52.521  60.167  72.90766  75.2390  80.6495 162.077   100
# z <- split1(idx, x) 325.681 351.658 383.13293 364.2215 384.9765 881.486   100
# > y
# [[1]]
# [,1]
# 2018-03-01    1
# 2018-03-02    2
# 2018-03-03    3
# 
# [[2]]
# [,1]
# 2018-03-04    4
# 2018-03-05    5
# 2018-03-06    6
# 2018-03-07    7
# 2018-03-08    8
# 2018-03-09    9
# 2018-03-10   10
# 
# [[3]]
# [,1]
# 2018-03-11   11
# 2018-03-12   12
# 2018-03-13   13
# 2018-03-14   14
# 2018-03-15   15

Answer 2

Consider creating an external, equal-length vector of Week Number with %U format for weekdays starting on Sunday. See ?strftime.

%U

Week of the year as decimal number (00–53) using Sunday as the first day 1 of the week (and typically with the first Sunday of the year as day 1 of week 1). The US convention.

week_num <- format(idx, "%U")
group <- unname(split(x, f = week_num))
group

[[1]]

2018-03-01 1
2018-03-02 2
2018-03-03 3

[[2]]

2018-03-04  4
2018-03-05  5
2018-03-06  6
2018-03-07  7
2018-03-08  8
2018-03-09  9
2018-03-10 10

[[3]]

2018-03-11 11
2018-03-12 12
2018-03-13 13
2018-03-14 14
2018-03-15 15