Incorrect behavior with dplyr's left_join?

Question

Surely this is not intended? Is this something that happens in other parts of dplyr\'s functionality and should I be concerned? I love the performance and hat

Answer 1

I posted something similar the other day. I think what you need to do is to have ORDER as numeric (or possibly the other way around). A has ORDER has integer. But B has ORDER as numeric. At the moment, dplyr asks you to have group-by variables in the same class. I received a comment from an SO user saying that this is something Hadley and his team has been working on now. This issue will be fixed in the future.

A$ORDER <- as.numeric(A$ORDER)
left_join(A,B, by = "ORDER")

     ORDER               COST   AREA
1 30305720                  0     NA
2 30334659                         0
3 30379936           11430.52   2339
4 30406397 20196.279999999999   2162
5 30407697                  0  23040
6 30431950           10445.99 475466

UPDATE After exchanging comments with thelatemail, I decided to add more observations here.

CASE 1: Treat ORDER as numeric

A$ORDER <- as.numeric(A$ORDER)

> left_join(A,B, by = "ORDER")
     ORDER               COST   AREA
1 30305720                  0     NA
2 30334659                         0
3 30379936           11430.52   2339
4 30406397 20196.279999999999   2162
5 30407697                  0  23040
6 30431950           10445.99 475466

> left_join(B,A, by = "ORDER")
Source: local data frame [5 x 3]

     ORDER   AREA               COST
1 30334659      0                   
2 30379936   2339           11430.52
3 30406397   2162 20196.279999999999
4 30407697  23040                  0
5 30431950 475466           10445.99

If you have ORDER as integer in both A and B, that works too.

CASE 2: Treat ORDER as integer and numeric

> left_join(A,B, by = "ORDER")
     ORDER               COST AREA
1 30305720                  0   NA
2 30334659                      NA
3 30379936           11430.52   NA
4 30406397 20196.279999999999   NA
5 30407697                  0   NA
6 30431950           10445.99   NA

> left_join(B,A, by = "ORDER")
Source: local data frame [5 x 3]

     ORDER   AREA               COST
1 30334659      0                   
2 30379936   2339           11430.52
3 30406397   2162 20196.279999999999
4 30407697  23040                  0
5 30431950 475466           10445.99

As suggested by thelatemail, integer/numeric combination does not work. But numeric/integer combination works.

Given these observations, it is safe to be consistent in group-by variable at the moment. Alternatively, merge() is the way to go. It can handle integer and numeric.

> merge(A,B, by = "ORDER", all = TRUE)
     ORDER               COST   AREA
1 30305720                  0     NA
2 30334659                         0
3 30379936           11430.52   2339
4 30406397 20196.279999999999   2162
5 30407697                  0  23040
6 30431950           10445.99 475466 

> merge(B,A, by = "ORDER", all = TRUE)
     ORDER   AREA               COST
1 30305720     NA                  0
2 30334659      0                   
3 30379936   2339           11430.52
4 30406397   2162 20196.279999999999
5 30407697  23040                  0
6 30431950 475466           10445.99

UPDATE2 (as of the 8th of November, 2014)

I am using a dev version of dplyr(dplyr_0.3.0.9000), which you can download from Github. The issue above is now solved.

left_join(A,B, by = "ORDER")
#     ORDER               COST   AREA
#1 30305720                  0     NA
#2 30334659                         0
#3 30379936           11430.52   2339
#4 30406397 20196.279999999999   2162
#5 30407697                  0  23040
#6 30431950           10445.99 475466