我有一个事件的数据帧,看起来像这样:
EVENT DATE LONG LAT TYPE
1 1/1/2000 23 45 A
2 2/1/2000 23 45 B
3 3/1/2000 23 45 B
3 5/2/2000 22 56 A
4 6/2/2000 19 21 A
我想把它折叠起来,这样在同一位置连续几天发生的任何事件(由LONG、LAT定义(都会折叠成一个单独的事件,其中包含开始和结束日期以及相关类型的串联列。
因此,上表将变为:
EVENT START-DATE END-DATE LONG LAT TYPE
1 1/1/2000 3/1/2000 23 45 ABB
2 5/2/2000 5/2/2000 22 56 A
3 6/2/2000 6/2/2000 19 21 A
任何关于如何最好地解决这一问题的建议都将不胜感激。
这是Ronak Shah解决方案的修改版本,将同一位置的非连续事件作为单独的事件周期。
# expanded data sample
df <- data.frame(
DATE = as.Date(c("2000-01-01", "2000-01-02", "2000-01-03", "2000-01-05",
"2000-02-05", "2000-02-06", "2000-02-07"), format = "%Y-%m-%d"),
LONG = c(23, 23, 23, 23, 22, 19, 22),
LAT = c(45, 45, 45, 45, 56, 21, 56),
TYPE = c("A", "B", "B", "A", "A", "B", "A")
)
library(dplyr)
df %>%
group_by(LONG, LAT) %>%
arrange(DATE) %>%
mutate(DATE.diff = c(1, diff(DATE))) %>%
mutate(PERIOD = cumsum(DATE.diff != 1)) %>%
ungroup() %>%
group_by(LONG, LAT, PERIOD) %>%
summarise(START_DATE = min(DATE),
END_DATe = max(DATE),
TYPE = paste(TYPE, collapse = "")) %>%
ungroup()
# A tibble: 5 x 6
LONG LAT PERIOD START_DATE END_DATe TYPE
<dbl> <dbl> <int> <date> <date> <chr>
1 19 21 0 2000-02-06 2000-02-06 B
2 22 56 0 2000-02-05 2000-02-05 A
3 22 56 1 2000-02-07 2000-02-07 A
4 23 45 0 2000-01-01 2000-01-03 ABB
5 23 45 1 2000-01-05 2000-01-05 A
编辑添加对"PERIOD"变量的解释。
为了简单起见,让我们考虑一些连续的&相同位置的非连续事件,因此我们可以跳过group_by(LONG, LAT)
&arrange(DATE)
步骤:
# sample dataset of 10 events at the same location.
# first 3 are on consecutive days, next 2 are on consecutive days,
# next 4 are on consecutive days, & last 1 is on its own.
df2 <- data.frame(
DATE = as.Date(c("2001-01-01", "2001-01-02", "2001-01-03",
"2001-01-05", "2001-01-06",
"2001-02-01", "2001-02-02", "2001-02-03", "2001-02-04",
"2001-04-01"), format = "%Y-%m-%d"),
LONG = rep(23, 10),
LAT = rep(45, 10),
TYPE = LETTERS[1:10]
)
作为中间步骤,我们创建一些辅助变量:
"DATE.diff"计算当前行的日期与;上一行的日期。由于第一行在"2001-01-01"之前没有日期,因此我们将差值默认为1。
"非连续"表示计算的日期差不是1(即与前一天不连续(,还是1(即前一天连续(。如果需要考虑数据集中同一位置的当天事件,可以在此处将计算从
DATE.diff != 1
更改为DATE.diff > 1
。"PERIOD"跟踪"非连续"变量中TRUE结果的数量。从第一行开始,每当一行与前一行不连续时,"PERIOD"将递增1。
由于辅助变量,"PERIOD"对于每组连续日期具有不同的值。
df2.intermediate <- df2 %>%
mutate(DATE.diff = c(1, diff(DATE))) %>%
mutate(non.consecutive = DATE.diff != 1) %>%
mutate(PERIOD = cumsum(non.consecutive))
> df2.intermediate
DATE LONG LAT TYPE DATE.diff non.consecutive PERIOD
1 2001-01-01 23 45 A 1 FALSE 0
2 2001-01-02 23 45 B 1 FALSE 0
3 2001-01-03 23 45 C 1 FALSE 0
4 2001-01-05 23 45 D 2 TRUE 1
5 2001-01-06 23 45 E 1 FALSE 1
6 2001-02-01 23 45 F 26 TRUE 2
7 2001-02-02 23 45 G 1 FALSE 2
8 2001-02-03 23 45 H 1 FALSE 2
9 2001-02-04 23 45 I 1 FALSE 2
10 2001-04-01 23 45 J 56 TRUE 3
然后,我们可以将"PERIOD"视为一个分组变量,以便找到开始/结束日期&每个周期内的事件:
df2.intermediate %>%
group_by(PERIOD) %>%
summarise(START_DATE = min(DATE),
END_DATe = max(DATE),
TYPE = paste(TYPE, collapse = "")) %>%
ungroup()
# A tibble: 4 x 4
PERIOD START_DATE END_DATe TYPE
<int> <date> <date> <chr>
1 0 2001-01-01 2001-01-03 ABC
2 1 2001-01-05 2001-01-06 DE
3 2 2001-02-01 2001-02-04 FGHI
4 3 2001-04-01 2001-04-01 J
使用dplyr
,我们可以按LAT
和LONG
进行分组,并为每组选择最大和最小DATE
,然后将TYPE
列粘贴在一起。
library(dplyr)
df %>%
group_by(LONG, LAT) %>%
summarise(start_date = min(as.Date(DATE, "%d/%m/%Y")),
end_date = max(as.Date(DATE, "%d/%m/%Y")),
type = paste0(TYPE, collapse = ""))
# LONG LAT start_date end_date type
# <int> <int> <date> <date> <chr>
#1 19 21 2000-02-06 2000-02-06 A
#2 22 56 2000-02-05 2000-02-05 A
#3 23 45 2000-01-01 2000-01-03 ABB