R - Modifying several interrelated columns with dynamic lag data simultaneously - r

Question

I have a dataset that look like
C_ID I_ID Loan R1 Prot_id Collateral R2 maxRank
1 A c 341 1 p1 506 1 3
2 A c 341 1 p2 366 2 3
3 A c 341 1 p3 263 3 3
4 A a 689 2 p1 506 1 3
5 A a 689 2 p2 366 2 3
6 A a 689 2 p3 263 3 3
7 A d 720 3 p1 506 1 3
8 A d 720 3 p2 366 2 3
9 A d 720 3 p3 263 3 3
10 A b 334 4 p1 506 1 3
11 A b 334 4 p2 366 2 3
12 A b 334 4 p3 263 3 3
13 A e 752 5 p1 506 1 3
14 A e 752 5 p2 366 2 3
15 A e 752 5 p3 263 3 3
16 B h 193 1 p5 529 1 2
17 B h 193 1 p4 414 2 2
18 B g 494 2 p5 529 1 2
19 B g 494 2 p4 414 2 2
20 B f 227 3 p5 529 1 2
21 B f 227 3 p4 414 2 2
22 B j 785 4 p5 529 1 2
23 B j 785 4 p4 414 2 2
24 B i 371 5 p5 529 1 2
25 B i 371 5 p4 414 2 2
26 B k 395 6 p5 529 1 2
27 B k 395 6 p4 414 2 2
Where R1 is ranking of loan for each contract_id group and R2 is ranking of each collateral under the cotnract_id. What is needed is
C_ID I_ID Loan R1 Prot_id Prot_value R2 maxRank PreAllocation Allocation PostAllocation Residual
1 A c 341 1 p1 506 1 3 341 341 0 165
2 A c 341 1 p2 366 2 3 0 0 0 366
3 A c 341 1 p3 263 3 3 0 0 0 263
4 A a 689 2 p1 506 1 3 689 165 524 0
5 A a 689 2 p2 366 2 3 524 366 158 0
6 A a 689 2 p3 263 3 3 158 158 0 105
7 A d 720 3 p1 506 1 3 720 0 720 0
8 A d 720 3 p2 366 2 3 720 0 720 0
9 A d 720 3 p3 263 3 3 720 105 615 0
10 A b 334 4 p1 506 1 3 334 0 334 0
11 A b 334 4 p2 366 2 3 334 0 334 0
12 A b 334 4 p3 263 3 3 334 0 334 0
13 A e 752 5 p1 506 1 3 752 0 752 0
14 A e 752 5 p2 366 2 3 752 0 752 0
15 A e 752 5 p3 263 3 3 752 0 752 0
16 B h 193 1 p5 529 1 2 193 193 0 336
17 B h 193 1 p4 414 2 2 0 0 0 414
18 B g 494 2 p5 529 1 2 494 336 158 0
19 B g 494 2 p4 414 2 2 158 158 0 256
20 B f 227 3 p5 529 1 2 227 0 227 0
21 B f 227 3 p4 414 2 2 227 227 0 29
22 B j 785 4 p5 529 1 2 785 0 785 0
23 B j 785 4 p4 414 2 2 785 29 756 0
24 B i 371 5 p5 529 1 2 371 0 371 0
25 B i 371 5 p4 414 2 2 371 0 371 0
26 B k 395 6 p5 529 1 2 395 0 395 0
27 B k 395 6 p4 414 2 2 395 0 395 0
Only the Allocation column is important and the other column are just to arrive at Allocation column. I was able to arrive at this using loop as below
df3 <- as.data.frame(df3)
df3$PreAllocation <- 0
df3$Allocation <- 0
df3$PostAllocation <- 0
df3$Residual <- 0
for (i in 1:nrow(df3)){
df3$PreAllocation[i] <- ifelse(df3$R2[i]==1,df3$Loan[i],df3$PostAllocation[i-1])
df3$Allocation[i]<- ifelse(df3$R1[i] >1, min(df3$Residual[i -
df3$maxRank[i]],df3$PreAllocation[i]),min(df3$PreAllocation[i],df3$Prot_value[i]))
df3$PostAllocation[i]<- df3$PreAllocation[i] - df3$Allocation[i]
df3$Residual[i] <- ifelse(df3$R1[i]==1, (df3$Prot_value[i] - df3$Allocation[i]), (df3$Residual[i-
df3$maxRank[i]] - df3$Allocation[i]))
}
However when dataset is big, there are performance issues. I have been trying to arrive at the same using apply functions; rowwise + transform etc but could not arrive at it. for
1. Columns are interdependent.
2. Need to use dynamic (based on maxRank) lag of columns being generated in calculation of later rows etc.
Any suggestion. Thanks.

Answer 1

It looks like a loop was used so you could look at the value in the previous row. Below is a solution using dplyr, which has a function lag() (and lead()) which let you look at previous (or successive) rows. It is also using pmin() (there's also a pmax()), which takes the min/max between corresponding elements in a set of vectors.
# `mutate()` takes data and one or more LHS=RHS statements about the data.
# Each column in the LHS is created (or overwritten) with the logic on the
# RHS. Conveniently, we don't have to prepend each with `df$3`.
# Uses dplyr's if_else()` instead of base R.
# `lag(x, n=1)` looks at previous row's value for x.
# Run `pmin(1:5, 5:1)` for a simple example
# of how it works.
df3 <- mutate(df3,
PreAllocation = if_else(R2 == 1, Loan, lag(PostAllocation, n = 1)),
Allocation = if_else(R3 > 1, pmin(lag(Residual, n = maxRank), PreAllocation),
pmin(PreAllocation, Prot_value)),
PostAllocation = PreAllocation - Allocation,
Residual = if_else(R1 == 1, Prot_value - Allocation, lag(Residual, n = maxRank - Allocation))
)
I encourage you to look at the dplyr CRAN page and the "Introduction to dplyr" vignette for further information.
If you'd like a syntax which is closer to base R's for subsetting & assignment, you might also consider the data.table package.
These are very popular frameworks for data manipulation and aggregation.