Add column to table with values depending on dates - r

Question

I have a table that has dates as a number and a value with each date. Now I'd like to add another column, weekSum, which contains the sum of value over the last week. However some dates are missing (so I can't always use the current and last 6 rows). My table looks like this:
df <- data.frame('date' = c(20160309, 20160310, 20160311, 20160312, 20160313, 20160314, 20160315, 20160317, 20160318, 20160319, 20160321), 'value' = c(1, 2, 3, 4, 5, 6, 7 ,8, 9, 10, 11))
date value
20160309 1
20160310 2
20160311 3
20160312 4
20160313 5
20160314 6
20160315 7
20160316 8
20160318 9 #17th skipped
20160319 10
20160321 11 #20th skipped
I'd like to get the following as output:
date value weekSum
20160309 1 NA
20160310 2 NA
20160311 3 NA
20160312 4 NA
20160313 5 NA
20160314 6 NA
20160315 7 28 # 1+2+3+4+5+6+7
20160316 8 35 # 2+3+4+5+6+7+8
20160318 9 39 # 4+5+6+7+8+9
20160319 10 45 # 5+6+7+8+9+10
20160321 11 45 # 7+8+9+10+11
How can this be done?

Answer 1

1) Convert the data frame to zoo and define a weekSum function which subsets its input to the last week and sums that. Then use rollapplyr with coredata = FALSE so that it passes a zoo object with times, not just the core data, to the weekSum function.
library(zoo)
z <- read.zoo(df, format = "%Y%m%d")
weekSum <- function(z) sum(z[time(z) > tail(time(z), 1) - 7])
transform(df, weekSum = rollapplyr(z, 7, weekSum, fill = NA, coredata = FALSE))
giving:
date value weekSum
2016-03-09 20160309 1 NA
2016-03-10 20160310 2 NA
2016-03-11 20160311 3 NA
2016-03-12 20160312 4 NA
2016-03-13 20160313 5 NA
2016-03-14 20160314 6 NA
2016-03-15 20160315 7 28
2016-03-16 20160316 8 35
2016-03-18 20160318 9 39
2016-03-19 20160319 10 45
2016-03-21 20160321 11 45
2) An alternative is to fill in the value at the missing dates with zero and then just use rollsumr with width of 7. z is from (1).
z0 <- merge(z, zoo(, seq(start(z), end(z), "day")), fill = 0)
transform(df, weekSum = rollsumr(z0, 7, fill = NA)[z0 != 0])

Answer 2

With base R it can be done like this:
res <- merge(df, data.frame(date = seq(df$date[1], to = df$date[length(d)], by = "days")), all.y = TRUE)
res$weekSum <- NA
for(i in seq_along(res$sum)[-seq_len(6)]){
res$weekSum[i] <- sum(res$value[(i - 6):i], na.rm = TRUE)
}
res <- res[!is.na(res$value), ]
res
# date value sum weekSum
#1 2016-03-09 1 NA NA
#2 2016-03-10 2 NA NA
#3 2016-03-11 3 NA NA
#4 2016-03-12 4 NA NA
#5 2016-03-13 5 NA NA
#6 2016-03-14 6 NA NA
#7 2016-03-15 7 28 28
#9 2016-03-17 8 33 35
#10 2016-03-18 9 39 42
#11 2016-03-19 10 45 49
#13 2016-03-21 11 45 56

Answer 3

Here is an approach using tidyverse tools. This method uses tidyr::complete to construct the full date sequence, making it easy to take the current row and the previous 6 as suggested. Be careful here if there are
NA values in value to begin with, as currently those rows will be filtered out at the end. Tweaks possible to avoid this case if necessary.
library(tidyverse)
library(lubridate)
#>
#> Attaching package: 'lubridate'
#> The following object is masked from 'package:base':
#>
#> date
df <- data.frame('date' = c(20160309, 20160310, 20160311, 20160312, 20160313, 20160314, 20160315, 20160317, 20160318, 20160319, 20160321), 'value' = c(1, 2, 3, 4, 5, 6, 7 ,8, 9, 10, 11))
df %>%
mutate(date = ymd(date)) %>%
complete(date = seq.Date(min(date), max(date), by = 1)) %>%
arrange(date) %>%
mutate(
newval = replace_na(value, 0),
weekSum = newval + lag(newval) + lag(newval, 2) + lag(newval, 3) +
lag(newval, 4) + lag(newval, 5) + lag(newval, 6)
) %>%
select(-newval) %>%
filter(!is.na(value))
#> # A tibble: 11 x 3
#> date value weekSum
#> <date> <dbl> <dbl>
#> 1 2016-03-09 1. NA
#> 2 2016-03-10 2. NA
#> 3 2016-03-11 3. NA
#> 4 2016-03-12 4. NA
#> 5 2016-03-13 5. NA
#> 6 2016-03-14 6. NA
#> 7 2016-03-15 7. 28.
#> 8 2016-03-17 8. 33.
#> 9 2016-03-18 9. 39.
#> 10 2016-03-19 10. 45.
#> 11 2016-03-21 11. 45.
Created on 2018-05-07 by the reprex package (v0.2.0).