Remove sequential repeating rows from xts object - r

Question

Say I have the following XTS object containing Bid and Ask data:
Time Bid Ask
00:01 10 11
00:02 10 11
00:03 11 12
00:04 12 13
00:05 10 11
00:06 10 11
00:07 10 11
00:08 9 12
00:09 2 10
00:10 4 5
I would like to get the following output:
Time Bid Ask
00:01 10 11
00:03 11 12
00:04 12 13
00:05 10 11
00:08 9 12
00:09 2 10
00:10 4 5
Rows should only be removed if they are the same as the previous entry. If there is a Bid or Ask change then nothing it is not removed, so simply taking out duplicates will not work.
This should be fairly simple as I have done this before, but I just cannot remember how and cannot find it.
Update:
I added some extra entries in my initial data and expected output.
Joshua's swearer is brilliant, but it depends on the function such as rowSums giving different results, but it breaks with 9 12.
I used the rowProds function from the matrixStats package which works, but obviously fails for my last 2 lines. Also, my sample the values in column one are smaller than in column 2. While that makes sense, it not a must so the function should work if the Ask was less than the Bid where both rowSums and rowProds would fail.
Is there a better row function that would always give a different result if anything is different, such as maybe a row hash?

Answer 1

You can do this by using rle() on the sum of each row.
x <- structure(
c(10L, 10L, 11L, 12L, 10L, 10L, 10L, 11L, 11L, 12L, 13L, 11L, 11L, 11L),
.Dim = c(7L, 2L), .Dimnames = list(NULL, c("Bid", "Ask")),
index = structure(1:7, tzone = "", tclass = c("POSIXct", "POSIXt")),
.indexCLASS = c("POSIXct", "POSIXt"), .indexTZ = "",
tclass = c("POSIXct", "POSIXt"), tzone = "", class = c("xts", "zoo"))
r <- rle(rowSums(x))
If you want the last observation in each group, you can just use cumsum(r$lengths) as the row index when subsetting.
R> x[cumsum(r$lengths),]
Bid Ask
1969-12-31 18:00:02 10 11
1969-12-31 18:00:03 11 12
1969-12-31 18:00:04 12 13
1969-12-31 18:00:07 10 11
Since you want the first observation of each group, you need to prepend the r$lengths vector with a 1 (you always want the first observation) and then remove the last element of r$lengths. Then call cumsum() on the result.
R> x[cumsum(c(1, head(r$lengths, -1))),]
Bid Ask
1969-12-31 18:00:01 10 11
1969-12-31 18:00:03 11 12
1969-12-31 18:00:04 12 13
1969-12-31 18:00:05 10 11
Good catch on the limitation of rowSums(). A robust solution is to diff() the bids and asks and select the rows where either is not zero.
d <- diff(x) != 0 # rows with price changes
d[1,] <- TRUE # always select first observation
g <- cumsum(d$Bid | d$Ask) # groups of repeats
r <- rle(as.numeric(g)) # run length encoding on groups
# now use the solution above
x[cumsum(c(1, head(r$lengths, -1))),]