Join two data frames in R based on closest timestamp

Question

Hi I have two tables (table1 and table2 below) and would like to join them based on the closest timestamp to form expected_output. Some kind of solution involving dplyr would be great if possible, but not if it further complicates things.

table1 = 
structure(list(date = structure(c(1437051300, 1434773700, 1431457200
), class = c("POSIXct", "POSIXt"), tzone = ""), val1 = c(94L, 
33L, 53L)), .Names = c("date", "val1"), row.names = c(NA, -3L
), class = "data.frame")

table2 = 
structure(list(date = structure(c(1430248288, 1435690482, 1434050843
), class = c("POSIXct", "POSIXt"), tzone = ""), val2 = c(67L, 
90L, 18L)), .Names = c("date", "val2"), row.names = c(NA, -3L
), class = "data.frame")

expected_output = 
structure(list(date = structure(c(1437051300, 1434773700, 1431457200
), class = c("POSIXct", "POSIXt"), tzone = ""), val1 = c(94L,
33L, 53L), val2 = c(90L, 18L, 67L)), .Names = c("date", "val1", 
"val2"), row.names = c(NA, -3L), class = "data.frame")

Lantana

Answer 1

Using rolling joins feature of data.table with roll = "nearest":

require(data.table) # v1.9.6+
setDT(table1)[, val2 := setDT(table2)[table1, val2, on = "date", roll = "nearest"]]

Here, val2 column is created by performing a join on the column date with roll = "nearest" option. For each row of table1$date, the closest matching row from table2$date is computed, and val2 for corresponding row is extracted.

Satori

This is very helpful! Any idea how to change from "nearest" to "most recent?" I.e. keep time directional, and not merge to a row in the future? – Picrite

@Picrite roll=Inf directional unlimited. roll=30 directional limit to staleness. Use sign to control direction. – Charlenacharlene

Answer 2

This is liable to be slow, but...

d   <- function(x,y) abs(x-y) # define the distance function
idx <- sapply( table1$date, function(x) which.min( d(x,table2$date) )) # find matches

cbind(table1,table2[idx,-1,drop=FALSE])
#                  date val1 val2
# 2 2015-07-16 08:55:00   94   90
# 3 2015-06-20 00:15:00   33   18
# 1 2015-05-12 15:00:00   53   67

Another way of constructing idx is max.col(-outer(table1$date, table2$date, d)).

Hardandfast