Fastest way to add rows for missing time steps?
I have a column in my datasets where time periods (Time) are integers ranging from a-b. Sometimes there might be missing time periods for any given group. I'd like to fill in those rows with NA. Below is example data for 1 (of several 1000) group(s).
structure(list(Id = c(1, 1, 1, 1), Time = c(1, 2, 4, 5), Value = c(0.568780482159894,
-0.7207749516298, 1.24258192959273, 0.682123081696789)), .Names = c("Id",
"Time", "Value"), row.names = c(NA, 4L), class = "data.frame")
Id Time Value
1 1 1 0.5687805
2 1 2 -0.7207750
3 1 4 1.2425819
4 1 5 0.6821231
As you can see, Time 3 is missing. Often one or more could be missing. I can solve this on my own but am afraid I wouldn't be doing this the most efficient way. My approach would be to create a function that:
Generate a sequence of time periods from min(Time) to max(Time)
Then do a setdiff to grab missing Time values.
Convert that vector to a data.frame
Pull unique identifier variables (Id and others not listed above), and add that to this data.frame.
Merge the two.
Return from function.
So the entire process would then get executed as below:
# Split the data into individual data.frames by Id.
temp_list <- dlply(original_data, .(Id))
# pad each data.frame
tlist2 <- llply(temp_list, my_pad_function)
# collapse the list back to a data.frame
filled_in_data <- ldply(tlist2)
Better way to achieve this?
Answer
Following up on comments with Ben Barnes and starting with his mydf3 :
DT = as.data.table(mydf3)
setkey(DT,Id,Time)
DT[CJ(unique(Id),seq(min(Time),max(Time)))]
Id Time Value Id2
[1,] 1 1 -0.262482283 2
[2,] 1 2 -1.423935165 2
[3,] 1 3 0.500523295 1
[4,] 1 4 -1.912687398 1
[5,] 1 5 -1.459766444 2
[6,] 1 6 -0.691736451 1
[7,] 1 7 NA NA
[8,] 1 8 0.001041489 2
[9,] 1 9 0.495820559 2
[10,] 1 10 -0.673167744 1
First 10 rows of 12800 printed.
setkey(DT,Id,Id2,Time)
DT[CJ(unique(Id),unique(Id2),seq(min(Time),max(Time)))]
Id Id2 Time Value
[1,] 1 1 1 NA
[2,] 1 1 2 NA
[3,] 1 1 3 0.5005233
[4,] 1 1 4 -1.9126874
[5,] 1 1 5 NA
[6,] 1 1 6 -0.6917365
[7,] 1 1 7 NA
[8,] 1 1 8 NA
[9,] 1 1 9 NA
[10,] 1 1 10 -0.6731677
First 10 rows of 25600 printed.
CJ stands for Cross Join, see ?CJ. The padding with NAs happens because nomatch by default is NA. Set nomatch to 0 instead to remove the no matches. If instead of padding with NAs the prevailing row is required, just add roll=TRUE. This can be more efficient than padding with NAs and then filling NAs afterwards. See the description of roll in ?data.table.
setkey(DT,Id,Time)
DT[CJ(unique(Id),seq(min(Time),max(Time))),roll=TRUE]
Id Time Value Id2
[1,] 1 1 -0.262482283 2
[2,] 1 2 -1.423935165 2
[3,] 1 3 0.500523295 1
[4,] 1 4 -1.912687398 1
[5,] 1 5 -1.459766444 2
[6,] 1 6 -0.691736451 1
[7,] 1 7 -0.691736451 1
[8,] 1 8 0.001041489 2
[9,] 1 9 0.495820559 2
[10,] 1 10 -0.673167744 1
First 10 rows of 12800 printed.
setkey(DT,Id,Id2,Time)
DT[CJ(unique(Id),unique(Id2),seq(min(Time),max(Time))),roll=TRUE]
Id Id2 Time Value
[1,] 1 1 1 NA
[2,] 1 1 2 NA
[3,] 1 1 3 0.5005233
[4,] 1 1 4 -1.9126874
[5,] 1 1 5 -1.9126874
[6,] 1 1 6 -0.6917365
[7,] 1 1 7 -0.6917365
[8,] 1 1 8 -0.6917365
[9,] 1 1 9 -0.6917365
[10,] 1 1 10 -0.6731677
First 10 rows of 25600 printed.
Instead of setting keys, you may use on. CJ also takes a unique argument. A small example with two 'Id':
d <- data.table(Id = rep(1:2, 4:3), Time = c(1, 2, 4, 5, 2, 3, 4), val = 1:7)
d[CJ(Id, Time = seq(min(Time), max(Time)), unique = TRUE), on = .(Id, Time)]
# Id Time val
# 1: 1 1 1
# 2: 1 2 2
# 3: 1 3 NA
# 4: 1 4 3
# 5: 1 5 4
# 6: 2 1 NA
# 7: 2 2 5
# 8: 2 3 6
# 9: 2 4 7
# 10: 2 5 NA
In this particular case, where one of the vectors in CJ was generated with seq, the result needs to be named explictly in order to match the names specified in on. When using bare variables in CJ though (like 'Id' here), they are auto-named, like in data.table() (from data.table 1.12.2).