问题
I'm just trying to figure out how to do a conditional join on two data.tables.
I've written a sqldf conditional join to give me the circuits whose start or finish times are within the other's start/finish times.
sqldf("select dt2.start, dt2.finish, dt2.counts, dt1.id, dt1.circuit
from dt2
left join dt1 on (
(dt2.start >= dt1.start and dt2.start < dt1.finish) or
(dt2.finish >= dt1.start and dt2.finish < dt1.finish)
)")
This gives me the correct result, but it's too slow for my large-ish data set.
What's the data.table way to do this without a vector scan?
Here's my data:
dt1 <- data.table(structure(list(circuit = structure(c(2L, 1L, 2L, 1L, 2L, 3L,
1L, 1L, 2L), .Label = c("a", "b", "c"), class = "factor"), start = structure(c(1393621200,
1393627920, 1393628400, 1393631520, 1393650300, 1393646400, 1393656000,
1393668000, 1393666200), class = c("POSIXct", "POSIXt"), tzone = ""),
end = structure(c(1393626600, 1393631519, 1393639200, 1393632000,
1393660500, 1393673400, 1393667999, 1393671600, 1393677000
), class = c("POSIXct", "POSIXt"), tzone = ""), id = structure(1:9, .Label = c("1001",
"1002", "1003", "1004", "1005", "1006", "1007", "1008", "1009"
), class = "factor")), .Names = c("circuit", "start", "end",
"id"), class = "data.frame", row.names = c(NA, -9L)))
dt2 <- data.table(structure(list(start = structure(c(1393621200, 1393624800, 1393626600,
1393627919, 1393628399, 1393632000, 1393639200, 1393646399, 1393650299,
1393655999, 1393660500, 1393666199, 1393671600, 1393673400), class = c("POSIXct",
"POSIXt"), tzone = ""), end = structure(c(1393624799, 1393626600,
1393627919, 1393628399, 1393632000, 1393639200, 1393646399, 1393650299,
1393655999, 1393660500, 1393666199, 1393671600, 1393673400, 1393677000
), class = c("POSIXct", "POSIXt"), tzone = ""), seconds = c(3599L,
1800L, 1319L, 480L, 3601L, 7200L, 7199L, 3900L, 5700L, 4501L,
5699L, 5401L, 1800L, 3600L), counts = c(1L, 1L, 0L, 1L, 2L, 1L,
0L, 1L, 2L, 3L, 2L, 3L, 2L, 1L)), .Names = c("start", "end",
"seconds", "counts"), row.names = c(1L, 3L, 4L, 5L, 6L, 7L, 8L,
9L, 10L, 11L, 12L, 13L, 14L, 15L), class = "data.frame"))
回答1:
Using non-equi joins:
ans = dt1[dt2, on=.(start <= end, end > start),
.(i.start, i.end, counts, id, circuit, cndn = i.start < x.start & i.end >= x.end),
allow.cartesian=TRUE
][!cndn %in% TRUE]
The condition start <= end, end >= start (note the >= on both cases) would check if two intervals overlap by any means. The open interval on one side is accomplished by end > start part (> instead of >=). But still it also picks up the intervals of type:
dt1: start=================end
dt2: start--------------------------------end ## start < start, end > end
and
dt1: start=================end
dt2: start----------end ## end == end
The cndn column is to check and remove these cases. Hopefully, those cases aren't a lot so that we don't materialise unwanted rows unnecessarily.
PS: the solution in this case is not as straightforward as I'd like to still, and that's because the solution requires an OR operation. It is possible to do two conditional joins, and then bind them together though.
Perhaps at some point, we'll have to think about the feasibility of extending joins to these kinds of operations in a more straightforward manner.
回答2:
No idea if this performs faster, but here's a shot at a data table method. I reshape dt1 and use findInterval to identify where the times in dt2 line up with times in dt1.
dt1 <- data.table(structure(list(circuit = structure(c(2L, 1L, 2L, 1L, 2L, 3L,
1L, 1L, 2L), .Label = c("a", "b", "c"), class = "factor"), start = structure(c(1393621200,
1393627920, 1393628400, 1393631520, 1393650300, 1393646400, 1393656000,
1393668000, 1393666200), class = c("POSIXct", "POSIXt"), tzone = ""),
end = structure(c(1393626600, 1393631519, 1393639200, 1393632000,
1393660500, 1393673400, 1393667999, 1393671600, 1393677000
), class = c("POSIXct", "POSIXt"), tzone = ""), id = structure(1:9, .Label = c("1001",
"1002", "1003", "1004", "1005", "1006", "1007", "1008", "1009"
), class = "factor")), .Names = c("circuit", "start", "end",
"id"), class = "data.frame", row.names = c(NA, -9L)))
dt2 <- data.table(structure(list(start = structure(c(1393621200, 1393624800, 1393626600,
1393627919, 1393628399, 1393632000, 1393639200, 1393646399, 1393650299,
1393655999, 1393660500, 1393666199, 1393671600, 1393673400), class = c("POSIXct",
"POSIXt"), tzone = ""), end = structure(c(1393624799, 1393626600,
1393627919, 1393628399, 1393632000, 1393639200, 1393646399, 1393650299,
1393655999, 1393660500, 1393666199, 1393671600, 1393673400, 1393677000
), class = c("POSIXct", "POSIXt"), tzone = ""), seconds = c(3599L,
1800L, 1319L, 480L, 3601L, 7200L, 7199L, 3900L, 5700L, 4501L,
5699L, 5401L, 1800L, 3600L), counts = c(1L, 1L, 0L, 1L, 2L, 1L,
0L, 1L, 2L, 3L, 2L, 3L, 2L, 1L)), .Names = c("start", "end",
"seconds", "counts"), row.names = c(1L, 3L, 4L, 5L, 6L, 7L, 8L,
9L, 10L, 11L, 12L, 13L, 14L, 15L), class = "data.frame"))
# > dt1
# circuit start end id
# 1: b 2014-02-28 16:00:00 2014-02-28 17:30:00 1001
# 2: a 2014-02-28 17:52:00 2014-02-28 18:51:59 1002
# 3: b 2014-02-28 18:00:00 2014-02-28 21:00:00 1003
# 4: a 2014-02-28 18:52:00 2014-02-28 19:00:00 1004
# 5: b 2014-03-01 00:05:00 2014-03-01 02:55:00 1005
# 6: c 2014-02-28 23:00:00 2014-03-01 06:30:00 1006
# 7: a 2014-03-01 01:40:00 2014-03-01 04:59:59 1007
# 8: a 2014-03-01 05:00:00 2014-03-01 06:00:00 1008
# 9: b 2014-03-01 04:30:00 2014-03-01 07:30:00 1009
# > dt2
# start end seconds counts
# 1: 2014-02-28 16:00:00 2014-02-28 16:59:59 3599 1
# 2: 2014-02-28 17:00:00 2014-02-28 17:30:00 1800 1
# 3: 2014-02-28 17:30:00 2014-02-28 17:51:59 1319 0
# 4: 2014-02-28 17:51:59 2014-02-28 17:59:59 480 1
# 5: 2014-02-28 17:59:59 2014-02-28 19:00:00 3601 2
# 6: 2014-02-28 19:00:00 2014-02-28 21:00:00 7200 1
# 7: 2014-02-28 21:00:00 2014-02-28 22:59:59 7199 0
# 8: 2014-02-28 22:59:59 2014-03-01 00:04:59 3900 1
# 9: 2014-03-01 00:04:59 2014-03-01 01:39:59 5700 2
# 10: 2014-03-01 01:39:59 2014-03-01 02:55:00 4501 3
# 11: 2014-03-01 02:55:00 2014-03-01 04:29:59 5699 2
# 12: 2014-03-01 04:29:59 2014-03-01 06:00:00 5401 3
# 13: 2014-03-01 06:00:00 2014-03-01 06:30:00 1800 2
# 14: 2014-03-01 06:30:00 2014-03-01 07:30:00 3600 1
## reshapes dt1 from wide to long
## puts start and end times into one column and sorts by time
## this is so that you can use findInterval later
dt3 <- dt1[,list(time = c(start,end)), by = "circuit,id"][order(time)]
dt3[,ntvl := seq_len(nrow(dt3))]
# circuit id time ntvl
# 1: b 1001 2014-02-28 16:00:00 1
# 2: b 1001 2014-02-28 17:30:00 2
# 3: a 1002 2014-02-28 17:52:00 3
# 4: b 1003 2014-02-28 18:00:00 4
# 5: a 1002 2014-02-28 18:51:59 5
# 6: a 1004 2014-02-28 18:52:00 6
# 7: a 1004 2014-02-28 19:00:00 7
# 8: b 1003 2014-02-28 21:00:00 8
# 9: c 1006 2014-02-28 23:00:00 9
# 10: b 1005 2014-03-01 00:05:00 10
# 11: a 1007 2014-03-01 01:40:00 11
# 12: b 1005 2014-03-01 02:55:00 12
# 13: b 1009 2014-03-01 04:30:00 13
# 14: a 1007 2014-03-01 04:59:59 14
# 15: a 1008 2014-03-01 05:00:00 15
# 16: a 1008 2014-03-01 06:00:00 16
# 17: c 1006 2014-03-01 06:30:00 17
# 18: b 1009 2014-03-01 07:30:00 18
## map interval to id
dt4 <- dt3[,list(ntvl = seq(from = min(ntvl), to = max(ntvl)-1), by = 1),by = "circuit,id"]
setkey(dt4, ntvl)
# circuit id ntvl
# 1: b 1001 1
# 2: a 1002 3
# 3: a 1002 4
# 4: b 1003 4
# 5: b 1003 5
# 6: b 1003 6
# 7: a 1004 6
# 8: b 1003 7
# 9: c 1006 9
# 10: c 1006 10
# 11: b 1005 10
# 12: c 1006 11
# 13: b 1005 11
# 14: a 1007 11
# 15: c 1006 12
# 16: a 1007 12
# 17: c 1006 13
# 18: a 1007 13
# 19: b 1009 13
# 20: c 1006 14
# 21: b 1009 14
# 22: c 1006 15
# 23: b 1009 15
# 24: a 1008 15
# 25: c 1006 16
# 26: b 1009 16
# 27: b 1009 17
# circuit id ntvl
## finds intervals in dt2
dt2[,`:=`(ntvl_start = findInterval(start, dt3[["time"]], rightmost.closed = FALSE),
ntvl_end = findInterval(end, dt3[["time"]], rightmost.closed = FALSE))]
# start end seconds counts ntvl_start ntvl_end
# 1: 2014-02-28 16:00:00 2014-02-28 16:59:59 3599 1 1 1
# 2: 2014-02-28 17:00:00 2014-02-28 17:30:00 1800 1 1 2
# 3: 2014-02-28 17:30:00 2014-02-28 17:51:59 1319 0 2 2
# 4: 2014-02-28 17:51:59 2014-02-28 17:59:59 480 1 2 3
# 5: 2014-02-28 17:59:59 2014-02-28 19:00:00 3601 2 3 7
# 6: 2014-02-28 19:00:00 2014-02-28 21:00:00 7200 1 7 8
# 7: 2014-02-28 21:00:00 2014-02-28 22:59:59 7199 0 8 8
# 8: 2014-02-28 22:59:59 2014-03-01 00:04:59 3900 1 8 9
# 9: 2014-03-01 00:04:59 2014-03-01 01:39:59 5700 2 9 10
# 10: 2014-03-01 01:39:59 2014-03-01 02:55:00 4501 3 10 12
# 11: 2014-03-01 02:55:00 2014-03-01 04:29:59 5699 2 12 12
# 12: 2014-03-01 04:29:59 2014-03-01 06:00:00 5401 3 12 16
# 13: 2014-03-01 06:00:00 2014-03-01 06:30:00 1800 2 16 17
# 14: 2014-03-01 06:30:00 2014-03-01 07:30:00 3600 1 17 18
## joins, by start time, then by end time
## the commented out lines may be a better alternative
## if there are many NA values
setkey(dt2, ntvl_start)
dt_ans_start <- dt4[dt2, list(start,end,counts,id,circuit),nomatch = NA]
# dt_ans_start <- dt4[dt2, list(start,end,counts,id,circuit),nomatch = 0]
# dt_ans_start_na <- dt2[!dt4]
setkey(dt2, ntvl_end)
dt_ans_end <- dt4[dt2, list(start,end,counts,id,circuit),nomatch = NA]
# dt_ans_end <- dt4[dt2, list(start,end,counts,id,circuit),nomatch = 0]
# dt_ans_end_na <- dt2[!dt4]
## bring them all together and remove duplicates
dt_ans <- unique(rbind(dt_ans_start, dt_ans_end), by = c("start", "id"))
dt_ans <- dt_ans[!(is.na(id) & counts > 0)]
dt_ans[,ntvl := NULL]
setkey(dt_ans,start)
# start end counts id circuit
# 1: 2014-02-28 16:00:00 2014-02-28 16:59:59 1 1001 b
# 2: 2014-02-28 17:00:00 2014-02-28 17:30:00 1 1001 b
# 3: 2014-02-28 17:30:00 2014-02-28 17:51:59 0 NA NA
# 4: 2014-02-28 17:51:59 2014-02-28 17:59:59 1 1002 a
# 5: 2014-02-28 17:59:59 2014-02-28 19:00:00 2 1002 a
# 6: 2014-02-28 17:59:59 2014-02-28 19:00:00 2 1003 b
# 7: 2014-02-28 19:00:00 2014-02-28 21:00:00 1 1003 b
# 8: 2014-02-28 21:00:00 2014-02-28 22:59:59 0 NA NA
# 9: 2014-02-28 22:59:59 2014-03-01 00:04:59 1 1006 c
# 10: 2014-03-01 00:04:59 2014-03-01 01:39:59 2 1006 c
# 11: 2014-03-01 00:04:59 2014-03-01 01:39:59 2 1005 b
# 12: 2014-03-01 01:39:59 2014-03-01 02:55:00 3 1006 c
# 13: 2014-03-01 01:39:59 2014-03-01 02:55:00 3 1005 b
# 14: 2014-03-01 01:39:59 2014-03-01 02:55:00 3 1007 a
# 15: 2014-03-01 02:55:00 2014-03-01 04:29:59 2 1006 c
# 16: 2014-03-01 02:55:00 2014-03-01 04:29:59 2 1007 a
# 17: 2014-03-01 04:29:59 2014-03-01 06:00:00 3 1006 c
# 18: 2014-03-01 04:29:59 2014-03-01 06:00:00 3 1007 a
# 19: 2014-03-01 04:29:59 2014-03-01 06:00:00 3 1009 b
# 20: 2014-03-01 06:00:00 2014-03-01 06:30:00 2 1006 c
# 21: 2014-03-01 06:00:00 2014-03-01 06:30:00 2 1009 b
# 22: 2014-03-01 06:30:00 2014-03-01 07:30:00 1 1009 b
# start end counts id circuit
来源：https://stackoverflow.com/questions/22680641/conditional-inequality-join-in-data-table