Efficient algorithm to calculate values in data.frame without loop - r

Question

Here is the situation where I got kinda stuck with R. I have data table with one row for each day, something like this:
Date = c(as.Date("2015-12-31"), as.Date("2016-01-01"));
Month1 = c("DEC", "JAN");
Year1 = c("15", "16");
Price1 = c(100, 110);
Month2 = c(NA_character_, NA_character_);
Year2 = c(NA_character_, NA_character_);
Price2 = c(NA_integer_, NA_integer_);
Month3 = c(NA_character_, NA_character_);
Year3 = c(NA_character_, NA_character_);
Price3 = c(NA_integer_, NA_integer_);
Month4 = c(NA_character_, NA_character_);
Year4 = c(NA_character_, NA_character_);
Price4 = c(NA_integer_, NA_integer_);
dataSample = data.frame(Date, Month1, Year1, Price1, Month2, Year2, Price2, Month3, Year3, Price3, Month4, Year4, Price4);
Which gives such a table:
Date Month1 Year1 Price1 Month2 Year2 Price2 Month3 Year3 Price3 Month4 Year4 Price4
1 2015-12-31 DEC 15 100 <NA> <NA> NA <NA> <NA> NA <NA> <NA> NA
2 2016-01-01 JAN 16 110 <NA> <NA> NA <NA> <NA> NA <NA> <NA> NA
Now I need to calculate all months and prices for each. For that I have 2 other data frames:
Date = c(as.Date("2015-12-31"), as.Date("2015-12-31"), as.Date("2015-12-31"), as.Date("2016-01-01"), as.Date("2016-01-01"), as.Date("2016-01-01"));
Month.Start = c("DEC", "JAN", "FEB", "JAN", "FEB", "MAR");
Year.Start = c("15", "16", "16", "16", "16", "16")
Month.End = c("JAN", "FEB", "MAR", "FEB", "MAR", "APR");
Year.End = c("16", "16", "16", "16", "16", "16")
Diff = c(10, 15, -15, 19, -20, -5);
diffsOneMonth = data.frame(Date, Month.Start, Year.Start, Month.End, Year.End, Diff)
Date = c(as.Date("2015-12-31"), as.Date("2016-01-01"));
Month.Start = c("DEC", "MAR");
Year.Start = c("15", "16")
Month.End = c("MAR", "JUN");
Year.End = c("16", "16")
Diff = c(11, 25);
diffsThreeMonth = data.frame(Date, Month.Start, Year.Start, Month.End, Year.End, Diff)
Which gives me these tables:
One month price differences
Date Month.Start Year.Start Month.End Year.End Diff
1 2015-12-31 DEC 15 JAN 16 10
2 2015-12-31 JAN 16 FEB 16 15
3 2015-12-31 FEB 16 MAR 16 -15
4 2016-01-01 JAN 16 FEB 16 19
5 2016-01-01 FEB 16 MAR 16 -20
6 2016-01-01 MAR 16 APR 16 -5
Three month price differences
Date Month.Start Year.Start Month.End Year.End Diff
1 2015-12-31 DEC 15 MAR 16 20
2 2016-01-01 MAR 16 JUN 16 25
Now I must fill dataSample data frame by using data from differences tables. I check what start/end months/years are available there and have to fill those months/years in dataSample. Then take difference of price and set calculated price in dataSample. So for example in dataSample we start with DEC 15, then in diffsOneMonth we have entry DEC 15 - JAN 16 with difference 10 so we add it to DEC 15 price and get JAN 16 price 110:
Date Month1 Year1 Price1 Month2 Year2 Price2 Month3 Year3 Price3 Month4 Year4 Price4
1 2015-12-31 DEC 15 100 JAN 16 110 <NA> <NA> NA <NA> <NA> NA
2 2016-01-01 JAN 16 110 <NA> <NA> NA <NA> <NA> NA <NA> <NA> NA
Now its possible to do next month and then next etc. If we use diffsOneMonth only we would get desirable result like this:
Date Month1 Year1 Price1 Month2 Year2 Price2 Month3 Year3 Price3 Month4 Year4 Price4
1 2015-12-31 DEC 15 100 JAN 16 110 FEB 16 125 MAR 16 110
2 2016-01-01 JAN 16 110 FEB 16 129 MAR 16 109 APR 16 104
However there is additional requirement that I must use wider month spread to calculate prices if its possible. So for 2015-12-31 there exists three month spread from DEC 15 to MAR 16 which should override price from one month difference. So DEC 15 price is 110 and DEC 15 - MAR 16 difference is 11 which makes MAR 16 price not 110 but 111:
Date Month1 Year1 Price1 Month2 Year2 Price2 Month3 Year3 Price3 Month4 Year4 Price4
1 2015-12-31 DEC 15 100 JAN 16 110 FEB 16 125 MAR 16 111
2 2016-01-01 JAN 16 110 FEB 16 129 MAR 16 109 APR 16 104
So for this sample it would be my final desirable output.
Real data is much more complex, with 6 and 12 month differences and 64 months forward for each date. Also some months can be missing. I tried to do it with a loop but it was very slow, however I am not sure how to approach such a problem without a loop. I have created few helper methods to be able to calculate next year/month:
nextContract = function(currentMonth, currentYear, length = 1,
years = c("10", "11", "12", "13", "14", "15", "16", "17", "18"),
months = c("JAN", "FEB", "MAR", "APR", "MAY", "JUN", "JUL", "AUG", "SEP", "OCT", "NOV", "DEC")) {
mIdx <- match(currentMonth, months)+length;
yDiff = ifelse(length(months) < mIdx, mIdx / length(months) - ifelse(mIdx %% length(months) == 0, 1, 0), 0);
return(data.frame(nextMonth(currentMonth, length, months), nextYear(currentYear, length = yDiff)))
}
nextYear = function(currentYear, length = 1, years = c("10", "11", "12", "13", "14", "15", "16", "17", "18")) {
return(years[match(currentYear, years)+length]);
}
nextMonth = function(currentMonth, length = 1, months = c("JAN", "FEB", "MAR", "APR", "MAY", "JUN", "JUL", "AUG", "SEP", "OCT", "NOV", "DEC")) {
mIdx <- match(currentMonth, months)+length;
return(months[ifelse(length(months) < mIdx, ifelse(mIdx %% length(months) != 0, mIdx %% length(months), length(months)), mIdx)]);
}
Example of usage could be:
> nextContract("DEC", "15")
nextMonth.currentMonth..length..months. nextYear.currentYear..length...yDiff.
1 JAN 16
or:
> nextContract("DEC", "15", length = 3)
nextMonth.currentMonth..length..months. nextYear.currentYear..length...yDiff.
1 MAR 16
This got to be pretty long question but I hope someone will take time to review it :)
Thanks in advance!
EDIT
A little bit of improvement on proposed solution and I got what I needed:
outrightAndForwardRows <- list("1" = diffsOneMonth, "3" = diffsThreeMonth) %>%
bind_rows(.id = "time_step") %>%
left_join(dataSample %>%
select(Date, Price1, Month1, Year1) ) %>%
mutate(Day.Start = 1) %>%
mutate(Day.End = 1) %>%
mutate(Outright.Day = 1) %>%
unite("Contract.Start", Day.Start, Month.Start, Year.Start) %>%
unite("Contract.End", Day.End, Month.End, Year.End) %>%
unite("Contract.Outright", Outright.Day, Month1, Year1) %>%
mutate(time_step = as.numeric(time_step),
Contract.Start =
Contract.Start %>%
parse_date_time("%d_%b_%y")) %>%
mutate(Contract.End =
Contract.End %>%
parse_date_time("%d_%b_%y")) %>%
mutate(Contract.Outright =
Contract.Outright %>%
parse_date_time("%d_%b_%y")) %>%
group_by(time_step, Date) %>%
arrange(Contract.End) %>%
mutate(Price = cumsum(Diff) + Price1) %>%
group_by(Date, Contract.End) %>%
slice(time_step %>% which.max) %>%
ungroup() %>%
select(-time_step, -Diff, -Contract.Start)
#### add outright and forward months to the same columns
outright <- outrightAndForwardRows %>% select(Date, Price=Price1, Contract=Contract.Outright) %>% unique
forwardMonths <- outrightAndForwardRows %>% select(Date, Contract=Contract.End, Price)
# join and sort rows
joined <- rbind(outright, forwardMonths) %>% arrange(Date, Contract)
# add contract sequence
joined = data.table(joined)
joined = joined[, Contract.seq:=seq(.N), by=Date];
dcast(joined, Date ~ Contract.seq, value.var=c("Price", "Contract"))

Answer 1

Something like this:
library(dplyr)
library(tidyr)
library(lubridate)
list(`1` = diffsOneMonth,
`3` = diffsThreeMonth) %>%
bind_rows(.id = "time_step") %>%
left_join(dataSample %>%
select(Date, Price1, Month1, Year1) ) %>%
mutate(Day.Start = 1) %>%
unite("Date.Start", Day.Start, Month.Start, Year.Start) %>%
mutate(time_step = as.numeric(time_step),
Date.Start =
Date.Start %>%
parse_date_time("%d_%b_%y")) %>%
group_by(time_step, Date) %>%
arrange(Date.Start) %>%
mutate(Price = cumsum(Diff) + Price1) %>%
group_by(Date, Date.Start) %>%
slice(time_step %>% which.max)