column with previous result - r

Question

I'm working with R
WHAT I HAVE:
ID_1 ID_2 Date x_1 y_2
1 12 3 2011-12-21 15 10
2 12 13 2011-12-22 50 40
3 3 12 2011-12-22 20 30
4 15 13 2011-12-23 30 20
...
and so on
TARGET:
ID_1 ID_2 Date x_1 y_2 XX_1 YY_2
1 12 3 2011-12-21 15 10 0 0
2 12 13 2011-12-22 50 40 15 0
3 3 12 2011-12-22 20 30 10 50
4 15 13 2011-12-23 30 20 0 40
...
and so on
I want to see in XX_1 and in YY_2 the values from the columns x_1 and y_2 corresponding to the previous values of ID_1 and ID1_2 in or "0" in case of no value is available before that date. I don't know how to handle the fact that different values could be in ID_1 and ID_2 (like IDs 3 and 12 in the example).
#Ekatef
ID1 AND ID2 (find match of the whole ID row, even if the order of IDs is switched):
ID_1 ID_2 Date x_1 y_2 XX_1 YY_2
1 12 3 2011-12-21 15 10 0 0
2 12 13 2011-12-22 50 40 0 0
3 3 12 2011-12-22 20 30 10 15
4 15 13 2011-12-23 30 20 0 0
5 12 13 2011-12-23 10 5 50 40

Answer 1

The OP has requested to copy the previous value for an ID (if any) to the appropriate new column.
This can solved by reshaping multiple columns simultaneously from wide to long format, finding the previous value by shifting / lagging, and reshaping back to wide format:
library(data.table)
setDT(DF)[, rn := .I]
long <- melt(DF, id.vars = c("rn", "Date"), measure.vars = patterns("^ID", "^x|y"),
value.name = c("ID", "value"))
long[order(Date), previous := shift(value, fill = 0), by = ID]
dcast(long, rn + Date ~ variable, value.var = c("ID", "value", "previous"))
rn Date ID_1 ID_2 value_1 value_2 previous_1 previous_2
1: 1 2011-12-21 12 3 15 10 0 0
2: 2 2011-12-22 12 13 50 40 15 0
3: 3 2011-12-22 3 12 20 30 10 50
4: 4 2011-12-23 15 13 30 20 0 40
Alternatively, the final call to dcast() can be replaced by an update while joining:
DF[long, on = .(rn),
c("XX_1", "YY_2") := .(previous[variable == 1L], previous[variable == 2L])][
, rn := NULL]
DF
ID_1 ID_2 Date x_1 y_2 XX_1 YY_2
1: 12 3 2011-12-21 15 10 0 0
2: 12 13 2011-12-22 50 40 15 0
3: 3 12 2011-12-22 20 30 10 50
4: 15 13 2011-12-23 30 20 0 40
which reproduces exactly OP's expected result.
Data
library(data.table)
DF <- fread(
"i ID_1 ID_2 Date x_1 y_2
1 12 3 2011-12-21 15 10
2 12 13 2011-12-22 50 40
3 3 12 2011-12-22 20 30
4 15 13 2011-12-23 30 20 ",
drop = 1L
)

Answer 2

If I understand you correctly, the target ID should be looked up from the left to the right and from the bottom to the top in all the rows strictly above the given ID value. I would write the function to find the coordinates of the preceded ID like that
# find the indices of the preceded ID value
# #id_matrix == your_data_frame[, c("ID_1", "ID_2")]
# [#i_of_row, #i_of_col] are the coordinates of the considered ID
# i_of_row > 1
FindPreviousID <- function(id_matrix, i_of_row, i_of_col) {
shorten_matrix <- id_matrix[1:(i_of_row - 1),,drop = FALSE]
rev_ind <- match(table = rev(t(shorten_matrix)),
x = ids[i_of_row,i_of_col], nomatch = NA_real_)
n_row_found <- floor((length(shorten_matrix) - rev_ind)/2) + 1
n_col_found <- (length(shorten_matrix) - rev_ind) %% ncol(shorten_matrix) + 1
return(c(row = n_row_found, col = n_col_found))
}
...and use it to calculate XX_1 and YY2
# emulate the original dataframe
ID_1 <- c(12,12,3,15,16,3)
ID_2<-c(3,13,12,13,17,15)
ids <- cbind(ID_1, ID_2) # IDs columns
x1 <- c(15, 50, 20, 30, 51, 60)
y2 <- c(10, 40, 30, 20, 53, 62)
vars <- cbind(x1, y2) # x&y columns
# assuming that the first XX_1 & YY_2 should be always 0
indices_XX <- sapply(FUN = function(i) FindPreviousID(id_matrix = ids, i_of_col = 1, i),
X = seq(along.with = ids[, 1])[-1])
indices_YY <- sapply(FUN = function(i) FindPreviousID(id_matrix = ids, i_of_col = 2, i),
X = seq(along.with = ids[, 1])[-1])
# construct XX and YY columns
XX_column <- c(NA, vars[t(indices_XX)])
XX_column[is.na(XX_column)] <- 0
YY_column <- c(NA, vars[t(indices_YY)])
YY_column[is.na(YY_column)] <- 0
Hope, that helps :)
Upd If you are interested to find a pair of IDs instead of the single ID, the function should be redesigned. One of the possible solutions looks like this
FindPreviousIDsPair <- function(id_matrix, i_of_row) {
shorten_matrix <- id_matrix[1:(i_of_row - 1),,drop = FALSE]
string_to_search_for <- id_matrix[i_of_row, ]
string_to_search_for_sorted <-
string_to_search_for[order(string_to_search_for)]
found_rows_boolean <- sapply(FUN = function(i) all(shorten_matrix[i,
order(shorten_matrix[i, ])] ==
string_to_search_for_sorted), X = 1:(i_of_row - 1))
found_row_n <- ifelse(any(found_rows_boolean),
max(which(found_rows_boolean)), NA_real_)
found_col_of_DI1 <- ifelse(any(found_rows_boolean),
match(string_to_search_for[1], shorten_matrix[found_row_n, ]), NA_real_)
found_col_of_DI2 <- ifelse(any(found_rows_boolean),
match(string_to_search_for[2], shorten_matrix[found_row_n, ]), NA_real_)
return(c(found_row_n, found_col_of_DI1, found_col_of_DI2))
}
Application of the redisigned look-up function to calculate XX and YY
indices_of_vars <- sapply(FUN = function(i) FindPreviousIDsPair(id_matrix =
ids, i), X = seq(along.with = ids[, 1])[-1])
indices_XX <- indices_of_vars[1:2, ]
indices_YY <- indices_of_vars[c(1, 3), ]
XX_column <- c(NA, vars[t(indices_XX)])
XX_column[is.na(XX_column)] <- 0
YY_column <- c(NA, vars[t(indices_YY)])
YY_column[is.na(YY_column)] <- 0