16

こんにちは、時刻が言うよりも少ないPOSIXct行から行を選択する効率的な方法を探しています（ミリ秒は必要ないことに注意してください。たとえば、使用できます）data.table12:00:00ITime

set.seed(1); N = 1e7;
DT = data.table(dts = .POSIXct(1e5*rnorm(N), tz="GMT"))
DT
                               dts
#       1: 1969-12-31 06:35:54.618925
#       2: 1970-01-01 05:06:04.332422
#     ---                           
# 9999999: 1970-01-03 00:37:00.035565
#10000000: 1969-12-30 08:30:23.624506

1 つの解決策 (ここでの問題は、N が大きい場合にキャストにコストがかかる可能性があることです)

f <- function(t, st, et) {time <- as.ITime(t); return(time>=as.ITime(st) & time<=as.ITime(et))}
P <- function(t, s) { #geekTrader solution
    ep <- .parseISO8601(s) 
    if(grepl('T[0-9]{2}:[0-9]{2}:[0-9]{2}/T[0-9]{2}:[0-9]{2}:[0-9]{2}', s)){
        first.time <- as.double(ep$first.time)
        last.time <- as.double(ep$last.time)-31449600
        SecOfDay <- as.double(t) %% 86400
        return(SecOfDay >= first.time & SecOfDay <= last.time )
    } else {
        return(t >= ep$first.time & t <= ep$last.time)    
    }
}

パフォーマンスについてのクイックルック

system.time(resf <- DT[f(dts,'00:00:00','11:59:59')])
   user  system elapsed 
   1.01    0.28    1.29
system.time(resP <- DT[P(dts,'T00:00:00/T11:59:59')])
   user  system elapsed 
   0.64    0.13    0.76 

identical(resf,resP)
[1] TRUE
4

4 に答える 4

7 
 P <- function(t, s) {
  ep <- .parseISO8601(s)

  if(grepl('T[0-9]{2}:[0-9]{2}:[0-9]{2}/T[0-9]{2}:[0-9]{2}:[0-9]{2}', s)){
    first.time <- as.double(ep$first.time)
    last.time <- as.double(ep$last.time)-31449600
    SecOfDay <- as.double(t) %% 86400
    return(SecOfDay >= first.time & SecOfDay <= last.time )

  } else {
    return(t >= ep$first.time & t <= ep$last.time)    
  }

}

F <- function(t, st, et) {
  time <- as.ITime(t) 
  return(time>=as.ITime(st) & time<=as.ITime(et))
}


 Sys.setenv(TZ='GMT')
 N = 1e7;
 set.seed(1);

 DT <- data.table(dts = .POSIXct(1e5*rnorm(N), tz="GMT"))


 system.time(resP <- DT[P(dts, 'T00:00:00/T12:00:00'), ])
##   user  system elapsed 
##   1.11    0.11    1.22 
 system.time(resF <- DT[F(dts,'00:00:00','12:00:00')])
##   user  system elapsed 
##   2.22    0.29    2.51 

 resP
##                         dts
##      1: 1969-12-31 06:35:54
##      2: 1970-01-01 05:06:04
##      3: 1969-12-31 00:47:17
##      4: 1970-01-01 09:09:10
##      5: 1969-12-31 01:12:33
##     ---                    
##5000672: 1970-01-01 06:08:15
##5000673: 1970-01-01 05:02:27
##5000674: 1969-12-31 02:25:24
##5000675: 1970-01-03 00:37:00
##5000676: 1969-12-30 08:30:23
 resF
##                         dts
##      1: 1969-12-31 06:35:54
##      2: 1970-01-01 05:06:04
##      3: 1969-12-31 00:47:17
##      4: 1970-01-01 09:09:10
##      5: 1969-12-31 01:12:33
##     ---                    
##5000672: 1970-01-01 06:08:15
##5000673: 1970-01-01 05:02:27
##5000674: 1969-12-31 02:25:24
##5000675: 1970-01-03 00:37:00
##5000676: 1969-12-30 08:30:23

 #Check the correctness
 resP[,list(mindts=max(dts)),by=list(as.Date(dts))]
##       as.Date              mindts
## 1: 1969-12-31 1969-12-31 12:00:00
## 2: 1970-01-01 1970-01-01 12:00:00
## 3: 1969-12-29 1969-12-29 12:00:00
## 4: 1970-01-02 1970-01-02 12:00:00
## 5: 1969-12-30 1969-12-30 12:00:00
## 6: 1970-01-03 1970-01-03 12:00:00
## 7: 1970-01-04 1970-01-04 11:59:59
## 8: 1970-01-05 1970-01-05 11:59:45
## 9: 1969-12-28 1969-12-28 12:00:00
##10: 1969-12-27 1969-12-27 11:59:21
##11: 1970-01-06 1970-01-06 10:53:21
##12: 1969-12-26 1969-12-26 10:15:03
##13: 1970-01-07 1970-01-07 08:21:55
 resF[,list(mindts=max(dts)),by=list(as.Date(dts))]
##       as.Date              mindts
## 1: 1969-12-31 1969-12-31 12:00:00
## 2: 1970-01-01 1970-01-01 12:00:00
## 3: 1969-12-29 1969-12-29 12:00:00
## 4: 1970-01-02 1970-01-02 12:00:00
## 5: 1969-12-30 1969-12-30 12:00:00
## 6: 1970-01-03 1970-01-03 12:00:00
## 7: 1970-01-04 1970-01-04 11:59:59
## 8: 1970-01-05 1970-01-05 11:59:45
## 9: 1969-12-28 1969-12-28 12:00:00
##10: 1969-12-27 1969-12-27 11:59:21
##11: 1970-01-06 1970-01-06 10:53:21
##12: 1969-12-26 1969-12-26 10:15:03
##13: 1970-01-07 1970-01-07 08:21:55

素敵なxtsスタイルのサブセット化のデモ

 DT[P(dts, '1970')]
##                         dts
##      1: 1970-01-01 05:06:04
##      2: 1970-01-02 20:18:48
##      3: 1970-01-01 09:09:10
##      4: 1970-01-01 13:32:22
##      5: 1970-01-01 20:30:32
##     ---                    
##5001741: 1970-01-02 15:51:12
##5001742: 1970-01-03 01:41:31
##5001743: 1970-01-01 06:08:15
##5001744: 1970-01-01 05:02:27
##5001745: 1970-01-03 00:37:00
 DT[P(dts, '197001')]
##                         dts
##      1: 1970-01-01 05:06:04
##      2: 1970-01-02 20:18:48
##      3: 1970-01-01 09:09:10
##      4: 1970-01-01 13:32:22
##      5: 1970-01-01 20:30:32
##     ---                    
##5001741: 1970-01-02 15:51:12
##5001742: 1970-01-03 01:41:31
##5001743: 1970-01-01 06:08:15
##5001744: 1970-01-01 05:02:27
##5001745: 1970-01-03 00:37:00
 DT[P(dts, '19700102')]
##                         dts
##      1: 1970-01-02 20:18:48
##      2: 1970-01-02 17:59:38
##      3: 1970-01-02 07:14:53
##      4: 1970-01-02 02:13:03
##      5: 1970-01-02 01:31:37
##     ---                    
##1519426: 1970-01-02 11:25:24
##1519427: 1970-01-02 10:00:21
##1519428: 1970-01-02 05:21:25
##1519429: 1970-01-02 05:11:26
##1519430: 1970-01-02 15:51:12
 DT[P(dts, '19700102 00:00:00/19700103 12:00:00')]
##                         dts
##      1: 1970-01-02 20:18:48
##      2: 1970-01-02 17:59:38
##      3: 1970-01-02 07:14:53
##      4: 1970-01-02 02:13:03
##      5: 1970-01-02 01:31:37
##     ---                    
##1785762: 1970-01-02 05:21:25
##1785763: 1970-01-02 05:11:26
##1785764: 1970-01-02 15:51:12
##1785765: 1970-01-03 01:41:31
##1785766: 1970-01-03 00:37:00

 #Check the correctness again
 DT[P(dts, '19700102 00:00:00/19700103 12:00:00'), max(dts)]
##[1] "1970-01-03 12:00:00 GMT"
 DT[P(dts, '19700102 00:00:00/19700103 12:00:00'), min(dts)]
##[1] "1970-01-02 00:00:00 GMT"
于 2013-04-05T12:07:29.010 に答える
0

as.POSIXlt遅いエントリですが、解決策はベクトルの名前付きリストを作成すると思います。そのうちの時間だけが必要です

列をキーにしてITime、バイナリ検索を使用して、午後 12 時までの時間をサブセット化します。

午後 12 時までに数60*60 *12 - 1秒あるため、午後 12 時までseq_len(43199)のすべてを返します (ただし、午後 12 時は含まれません)。

# create IDate and ITime columns and key by time
setkey(DT[, c('Date','Time') := IDateTime(dts)],Time)

# subset times before 12pm
DT[.(seq_len(43199))]
于 2013-04-16T02:27:41.147 に答える