Joining/matching data frames in R

Question

I have two data frames.

The first one has two columns: x is water depth, y is temperature at each depth.

The second one has two columns too, x is also water depth, but at different depth compared to that in the first table. The second column z is salinity.

I want to join the two tables by x, by adding z to the first table. I have learned how to join tables using 'key' in tidyr , but that only works if the keys are identical. The x in these two tables are not the same.

What I want to do is to match the depth x in table 2 to that within 10% of that in table 1 (i.e. match 1.1 in table 2 x to 1.0 in table 1 x).

How can I do this?

Table 1
| x | y  |
|---|----|
| 1 | 25 |
| 2 | 26 |
| 3 | 27 |

Table 2
| x    | z  |
|------|----|
| 1.1  | 30 |
| 2.05 | 35 |
| 3.8  | 34 |

I want

Table 1
| x | y  | z  |
|---|----|----|
| 1 | 25 | 30 |
| 2 | 26 | 35 |
| 3 | 27 | NA |

The first two 'x' values have matches (within 10% absolute difference) in Table 2. The third one doesn't.

Answer 1

Edit after data was provided:

Taking data definition from @MKR's post:

library(fuzzyjoin)
fuzzy_left_join(Table_1, Table_2,match_fun = function(x,y)  y> x & y<=1.1*x )
# Joining by: "x"
#   x.x  y  x.y  z
# 1   1 25 1.10 30
# 2   2 26 2.05 35
# 3   3 27   NA NA

---

general explanations with fake data (first answer)

fake data

iris1 <- head(iris[1:2])
iris1
#   Sepal.Length Sepal.Width
# 1          5.1         3.5
# 2          4.9         3.0
# 3          4.7         3.2
# 4          4.6         3.1
# 5          5.0         3.6
# 6          5.4         3.9

iris2 <- head(iris[c(1,3)])
set.seed(1)

# add noise
iris2$Sepal.Length <- iris2$Sepal.Length + rnorm(6,sd=0.05)

# shuffle rows
iris2 <- iris2[sample(seq(nrow(iris2))),]

iris2
#   Sepal.Length Petal.Length
# 5     5.016475          1.4
# 2     4.909182          1.4
# 4     4.679764          1.5
# 6     5.358977          1.7
# 3     4.658219          1.3
# 1     5.068677          1.4

code

library(fuzzyjoin)
fuzzy_left_join(iris1,iris2,match_fun= function(x,y) y>0.99*x & y<1.01*x )
# Joining by: "Sepal.Length"
# Sepal.Length.x Sepal.Width Sepal.Length.y Petal.Length
# 1            5.1         3.5       5.068677          1.4
# 2            4.9         3.0       4.909182          1.4
# 3            4.7         3.2       4.679764          1.5
# 4            4.7         3.2       4.658219          1.3
# 5            4.6         3.1             NA           NA
# 6            5.0         3.6       5.016475          1.4
# 7            5.4         3.9       5.358977          1.7

We see that some rows matched really well, let's take a look at the exceptions. The row number 4 had too much noise added in iris2, so it got paired with row 3, which has 2 matches. As I chose a left join, row 4 is still shown, but with NAs for iris2's columns.

As I understand it:

• The joining columns will be expanded
• The function takes these long columns (6*6==36 elements here) as arguments
• We an apply vectorized functions (such as < or & in this case) to return a vector of logical that will filter these long columns in order to build the output data.frame.

distance_left_join is more straightforward to use, but then it's absolute distance, not relative.

Answer 2

An option using sqldf can be achieved as:

library(sqldf)


sqldf("select T1.x, T1.y, A.z from Table_1 T1
left join (select Table_1.x, Table_1.y, Table_2.z from Table_1 
   left join Table_2 where round((100*abs(Table_1.x - Table_2.x)/Table_1.x),0) <= 10) A 
on T1.x = A.x")

#   x  y  z
# 1 1 25 30
# 2 2 26 35
# 3 3 27 NA

Data:

Table_1 <- read.table(text = 
"x  y  
1   25  
2   26  
3   27",
header = TRUE)


Table_2 <- read.table(text = 
"x  z  
1.1    30  
2.05   35  
3.8    34",
header = TRUE)