Question

我有一个数据框，显示何时在不同位置检测到动物。我想从仅站点A的检测文件（df）中消除行（过滤器），如果在一个时间段（5分钟）内未在站点B上检测到单个动物。 我需要遍历每只动物和多个站点。我的真实数据有许多动物和超过一百万个检测观测值。我正在寻找一个高效的data.table解决方案。

这两个变量是个体（动物）和检测到的位置。

示例：

obs.num<-1:21 # a simple observation number 
animal<-c(rep("RBT 1",10),rep("RBT 2",7) ,rep("RBT 3",2),"RBT 4","RBT 2") # 
a fake list of animal id's (my data has many)
now <- Sys.time()
ts <- seq(from = now, length.out = 16, by = "mins")
ts <- c(ts,seq(from=tail(ts,1), length.out = 3, by = "hour")) # create a 
fake series of time stamps 
ts <- c(ts,seq(from=tail(ts,1), length.out = 2, by = "hour"))
df<-data.frame(obs.num,animal,ts) # make data frame 
df$site<-c("A","B","A","B","A","B","A","B","A","B","A","B","A","B","A","B","A","B","A","B","B")# make a fake series of sites detection occurred at 
str(df)
df # my example data frame

在此示例中，我想删除整个行以进行观察19。

我正在寻找类似于此解决方案的data.table解决方案

library(sqldf)

sqldf("with B as (select * from df where site == 'B')
  select distinct df.* from df 
  join B on df.animal = B.animal and 
        B.ts - df.ts between -5 * 60 and 5 * 60
  order by 1")

Answer 1

有点笨拙，但是您可以通过data.table中的非等参来实现：

library(data.table)
setDT(df)
nm = names(df)
# unfortunately non-equi-joins don't support on-the-fly
#   columns yet, so we have to first define them explicitly; see:
#   https://github.com/Rdatatable/data.table/issues/1639
df[ , ts_minus_5 := ts - 5*60]
df[ , ts_plus_5 := ts + 5*60]

# identify the observations _matching_ your criteria (i.e. those to keep)
found_at_b = unique(
  df[site == 'A'][df[site == 'B'], .(x.obs.num, x.animal),
                  on = .(animal == animal, ts >= ts_minus_5, ts <= ts_plus_5),
                  # allow.cartesian allows this join to return any
                  #   number of rows, necessary since any "B" row
                  #   might match multiple "A" rows;
                  # nomatch = 0L drops any "B" row without a 
                  #   match found in "A" rows
                  allow.cartesian = TRUE, nomatch = 0L]
)

# to filter, define a "drop" flag (could also call it "filter")
df[site == 'B', drop := FALSE]
df[found_at_b, on = c(obs.num = 'x.obs.num', animal = 'x.animal'),
   drop := FALSE]

# could define drop = TRUE for the other rows, but no need
df = df[(!drop)]

还有其他一些方法可以通过更加谨慎地潜在地创建副本来清理代码，也许首先split site-将数据[]放在一个{ {1}}通话等等，但这会让您入门。

基于posixct datetime和两个变量的data.table过滤解决方案？

1 个答案: