dplyr group by, carry forward value from previous group to next
This kind of use of first and last is very untidy, so we'll keep it for the latest step.
First we build intermediate data, following your code, but adding some columns to join later at the right places. I'm not sure if you need to keep all columns, you won't need the second join if not.
library(dplyr)
library(tidyr)
df1 <- df0 %>%
dplyr::mutate(RunID = data.table::rleid(x.long)) %>%
group_by(RunID) %>%
mutate(RunID_f = ifelse(row_number()==1,RunID,NA)) %>% # for later merge
mutate(RunID_l = ifelse(row_number()==n(),RunID,NA)) # possibly unneeded
Then we build summarized data, I refactored your code a bit as you see, because these operations "should" be rowwise.
summarized_data <- df1 %>%
filter(x.long !=0) %>%
summarize_at(vars(close.x,inital.capital),c("first","last")) %>%
mutate(x.long.share = inital.capital_first / close.x_first,
x.end.value = x.long.share * close.x_last,
x.net.profit = inital.capital_last - x.end.value,
new.initial.capital = x.net.profit + inital.capital_last,
lagged.new.initial.capital = lag(new.initial.capital,1))
# A tibble: 2 x 10
# RunID close.x_first inital.capital_first close.x_last inital.capital_last x.long.share x.end.value x.net.profit new.initial.capital lagged.new.initial.capital
# <int> <dbl> <int> <dbl> <int> <dbl> <dbl> <dbl> <dbl> <dbl>
# 1 3 38.85 10000 38.13 10000 257.4003 9814.672 185.3282 10185.328 NA
# 2 5 33.03 10000 34.34 10000 302.7551 10396.609 -396.6091 9603.391 10185.33
Then we join our summarized table to the original, getting advantage of the trick of the firt step. The first join may be skipped if you don't need all columns.
df2 <- df1 %>% ungroup %>%
left_join(summarized_data %>% select(-lagged.new.initial.capital) ,by=c("RunID_l"="RunID")) %>% # if you want the other variables, if not, skip the line
left_join(summarized_data %>% select(RunID,lagged.new.initial.capital) ,by=c("RunID_f"="RunID")) %>%
mutate(inital.capital = ifelse(is.na(lagged.new.initial.capital),inital.capital,lagged.new.initial.capital)) %>%
select(close.x:inital.capital) # for readability here
# # A tibble: 20 x 6
# close.x x.long y.short x.short y.long inital.capital
# <dbl> <int> <int> <int> <int> <dbl>
# 1 37.9600 NA NA NA NA 10000.00
# 2 36.5200 0 0 0 0 10000.00
# 3 38.3200 0 0 0 0 10000.00
# 4 38.5504 0 0 0 0 10000.00
# 5 38.1700 0 0 0 0 10000.00
# 6 38.8500 1 1 0 0 10000.00
# 7 38.5300 1 1 0 0 10000.00
# 8 39.1300 1 1 0 0 10000.00
# 9 38.1300 1 1 0 0 10000.00
# 10 37.0100 0 0 1 1 10000.00
# 11 36.1400 0 0 1 1 10000.00
# 12 35.2700 0 0 1 1 10000.00
# 13 35.1300 0 0 1 1 10000.00
# 14 32.2000 0 0 1 1 10000.00
# 15 33.0300 1 1 0 0 10185.33
# 16 34.9400 1 1 0 0 10000.00
# 17 34.5700 1 1 0 0 10000.00
# 18 33.6000 1 1 0 0 10000.00
# 19 34.3400 1 1 0 0 10000.00
# 20 35.8600 0 0 1 1 10000.00
data
df<- read.table(text="close.x x.long y.short x.short y.long inital.capital x.long.shares x.end.value x.net.profit new.initial.capital
37.96 NA NA NA NA 10000 NA NA NA NA
36.52 0 0 0 0 10000 0 0 0 0
38.32 0 0 0 0 10000 0 0 0 0
38.5504 0 0 0 0 10000 0 0 0 0
38.17 0 0 0 0 10000 0 0 0 0
38.85 1 1 0 0 10000 0 0 0 0
38.53 1 1 0 0 10000 0 0 0 0
39.13 1 1 0 0 10000 0 0 0 0
38.13 1 1 0 0 10000 257.4002574 9814.671815 185.3281853 10185.32819
37.01 0 0 1 1 10000 0 0 0 0
36.14 0 0 1 1 10000 0 0 0 0
35.27 0 0 1 1 10000 0 0 0 0
35.13 0 0 1 1 10000 0 0 0 0
32.2 0 0 1 1 10000 0 0 0 0
33.03 1 1 0 0 10000 0 0 0 0
34.94 1 1 0 0 10000 0 0 0 0
34.57 1 1 0 0 10000 0 0 0 0
33.6 1 1 0 0 10000 0 0 0 0
34.34 1 1 0 0 10000 302.7550711 10396.60914 -396.6091432 9603.390857
35.86 0 0 1 1 10000 0 0 0 0",stringsAsFactors=FALSE,header=TRUE)
df0 <- df %>% select(close.x:inital.capital)
You're using data.table in the question and have tagged the question data.table, so here is a data.table answer. When j evaluates, it's in a static scope where local variables retain their values from the previous group.
Using dummy data to demonstrate :
require(data.table)
set.seed(1)
DT = data.table( long = rep(c(0,1,0,1),each=3),
val = sample(5,12,replace=TRUE))
DT
long val
1: 0 2
2: 0 2
3: 0 3
4: 1 5
5: 1 2
6: 1 5
7: 0 5
8: 0 4
9: 0 4
10: 1 1
11: 1 2
12: 1 1
DT[, v1:=sum(val), by=rleid(long)][]
long val v1
1: 0 2 7
2: 0 2 7
3: 0 3 7
4: 1 5 12
5: 1 2 12
6: 1 5 12
7: 0 5 13
8: 0 4 13
9: 0 4 13
10: 1 1 4
11: 1 2 4
12: 1 1 4
So far, simple enough.
prev = NA # initialize previous group value
DT[, v2:={ans<-last(val)/prev; prev<-sum(val); ans}, by=rleid(long)][]
long val v1 v2
1: 0 2 7 NA
2: 0 2 7 NA
3: 0 3 7 NA
4: 1 5 12 0.71428571
5: 1 2 12 0.71428571
6: 1 5 12 0.71428571
7: 0 5 13 0.33333333
8: 0 4 13 0.33333333
9: 0 4 13 0.33333333
10: 1 1 4 0.07692308
11: 1 2 4 0.07692308
12: 1 1 4 0.07692308
> 3/NA
[1] NA
> 5/7
[1] 0.7142857
> 4/12
[1] 0.3333333
> 1/13
[1] 0.07692308
> prev
[1] NA
Notice that the prev value did not update because prev and ans are local variables inside j's scope that were being updated as each group ran. Just to illustrate, the global prev can be updated from within each group using R's <<- operator :
DT[, v2:={ans<-last(val)/prev; prev<<-sum(val); ans}, by=rleid(long)]
prev
[1] 4
But there's no need to use <<- in data.table as local variables are static (retain their values from previous group). Unless you need to use the final group's value after the query has finished.
You're going to have a hard time finding an 'elegant' pure-dplyr solution, because dplyr isn't really designed to do this. What dplyr likes to do is map/reduce type operations (mutate and summarize) that use window and summary functions respectively. What you're asking for isn't really either of those, because you want each group to depend on the last, so you're really describing a looping operation with side effects - two very non-R-philosophy operations.
If you want to hack your way into doing what you describe, you can try an approach like this:
new.initial.capital <- 0
for (z in split(df, df$x.long)) {
z$initial.capital[[1]] <- new.initial.capital
# some other calculations here
# maybe you want to modify df as well
new.initial.capital <- foo
}
However, this is really not a very R-friendly piece of code, as it depends on side effects and loops. I would advise seeing if you can reframe your calculations in terms of a summary and/or window function if you want to integrate with dplyr.
For more:
https://www.rstudio.com/wp-content/uploads/2015/02/data-wrangling-cheatsheet.pdf
https://danieljhocking.wordpress.com/2014/12/03/lags-and-moving-means-in-dplyr/