我在 R 的 tibble 中有一个数据集,如下所示:
# A tibble: 50,045 x 5
ref_key start_date end_date
<chr> <date> <date>
1 123 2010-01-08 2010-01-13
2 123 2010-01-21 2010-01-23
3 123 2010-03-10 2010-04-14
我需要创建另一个每行只存储一个日期的 tibble,如下所示:
ref_key date
<chr> <date>
1 123 2010-01-08
2 123 2010-01-09
3 123 2010-01-10
4 123 2010-01-11
5 123 2010-01-12
6 123 2010-01-13
7 123 2010-01-21
8 123 2010-01-22
9 123 2010-01-23
目前,我正在为此编写一个显式循环,如下所示:
for (loop in (1:nrow(input.df))) {
if (loop%%100==0) {
print(paste(loop,'/',nrow(input.df)))
}
temp.df.st00 <- input.df[loop,] %>% data.frame
temp.df.st01 <- tibble(ref_key=temp.df.st00[,'ref_key'],
date=seq(temp.df.st00[,'start_date'],
temp.df.st00[,'end_date'],1))
if (loop==1) {
output.df <- temp.df.st01
} else {
output.df <- output.df %>%
bind_rows(temp.df.st01)
}
}
它正在工作,但速度很慢,因为我有>50k 行要处理,完成循环需要几分钟。
我想知道这一步是否可以矢量化,是否与dplyr
中的row_wise
有关?
我们创建一个行名列(rownames_to_column
),然后nest
"rn"和"ref_key",mutate
在map
中获取"start_date"和"end_date"的序列,并在select
掉不需要的列后unnest
library(tidyverse)
res <- df1 %>%
rownames_to_column('rn') %>%
nest(-rn, -ref_key) %>%
mutate(date = map(data, ~ seq(.x$start_date, .x$end_date, by = "1 day"))) %>%
select(-data, -rn) %>%
unnest
head(res, 9)
# ref_key date
#1 123 2010-01-08
#2 123 2010-01-09
#3 123 2010-01-10
#4 123 2010-01-11
#5 123 2010-01-12
#6 123 2010-01-13
#7 123 2010-01-21
#8 123 2010-01-22
#9 123 2010-01-23
一种解决方案是使用tidyr::complete
来扩展行。由于行扩展基于行的start-date
和end_date
,因此group_by
row_number
将有助于生成start-date
和end_date
之间的Date
序列。
library(dplyr)
library(tidyr)
df %>% #mutate(rnum = row_number()) %>%
group_by(row_number()) %>%
complete(start_date = seq.Date(max(start_date), max(end_date), by="day")) %>%
fill(ref_key) %>%
ungroup() %>%
select(ref_key, date = start_date)
# # A tibble: 45 x 2
# ref_key date
# <int> <date>
# 1 123 2010-01-08
# 2 123 2010-01-09
# 3 123 2010-01-10
# 4 123 2010-01-11
# 5 123 2010-01-12
# 6 123 2010-01-13
# 7 123 2010-01-21
# 8 123 2010-01-22
# 9 123 2010-01-23
# 10 123 2010-03-10
# # ... with 35 more rows
数据
df <- read.table(text = "ref_key start_date end_date
123 2010-01-08 2010-01-13
123 2010-01-21 2010-01-23
123 2010-03-10 2010-04-14", header = TRUE, stringsAsFactor = FALSE)
df$start_date <- as.Date(df$start_date)
df$end_date <- as.Date(df$end_date)