data.table inner/outer join with NA in join column of type double bug?

Question

Following this wikipedia article SQL join I wanted to have a clear view on how we could have joins with data.table. In the process we might have uncovered a bug when joining with NAs. Taking the wiki example:

R) X = data.table(name=c("Raf","Jon","Ste","Rob","Smi","Joh"),depID=c(31,33,33,34,34,NA),key="depID")
R) Y = data.table(depID=c(31,33,34,35),depName=c("Sal","Eng","Cle","Mar"),key="depID")
R) X
   name depID
1:  Joh    NA
2:  Raf    31
3:  Jon    33
4:  Ste    33
5:  Rob    34
6:  Smi    34
R) Y
   depID depName
1:    31     Sal
2:    33     Eng
3:    34     Cle
4:    35     Mar

---

LEFT OUTER JOIN

R) merge.data.frame(X,Y,all.x=TRUE)
  depID name depName
1    31  Raf     Sal
2    33  Jon     Eng
3    33  Ste     Eng
4    34  Rob     Cle
5    34  Smi     Cle
6    NA  Joh    <NA>

merge.data.table do not output the same result and show what I think is a bug on lign 2.

R) merge(X,Y,all.x=TRUE)
   depID name depName
1:    NA  Joh     Eng
2:    31  Raf      NA
3:    33  Jon     Eng
4:    33  Ste     Eng
5:    34  Rob     Cle
6:    34  Smi     Cle
R) Y[X] #same -> :(
   depID depName name
1:    NA     Eng  Joh
2:    31      NA  Raf
3:    33     Eng  Jon
4:    33     Eng  Ste
5:    34     Cle  Rob
6:    34     Cle  Smi

---

RIGHT OUTER JOIN Looks like the same

R) merge.data.frame(X,Y,all.y=TRUE)
  depID name depName
1    31  Raf     Sal
2    33  Jon     Eng
3    33  Ste     Eng
4    34  Rob     Cle
5    34  Smi     Cle
6    35 <NA>     Mar

R) merge(X,Y,all.y=TRUE)
   depID name depName
1:    NA  Joh     Eng
2:    31   NA     Sal
3:    33  Jon     Eng
4:    33  Ste     Eng
5:    34  Rob     Cle 
6:    34  Smi     Cle
7:    35   NA     Mar

---

INNER (NATURAL) JOIN

R) merge.data.frame(X,Y)
  depID name depName
1    31  Raf     Sal
2    33  Jon     Eng
3    33  Ste     Eng
4    34  Rob     Cle
5    34  Smi     Cle
R) merge(X,Y)
   depID name depName
1:    NA  Joh     Eng
2:    33  Jon     Eng
3:    33  Ste     Eng
4:    34  Rob     Cle
5:    34  Smi     Cle

Grizzled

Answer 1

Yes it looks like an (embarassing) new bug related to the NA in key. There have been other discussions about NA in key not being possible but I didn't realise it could mess up in that way. Will investigate. Thanks ...

#2453 NA in double key column messes up joins (NA in integer and character ok)

Now fixed in 1.8.7 (commit 780), from NEWS :

NA in a join column of type double could cause both X[Y] and merge(X,Y) to return incorrect results, #2453. Due to an errant x==NA_REAL in the C source which should have been ISNA(x). Support for double in keyed joins is a relatively recent addition to data.table, but embarassing all the same. Fixed and tests added. Many thanks to statquant for the thorough and reproducible report.

Pachysandra

As was reported in a prior, deleted answer (which was really a comment), the merge works correctly if the depID columns are integers. – Caffey

@MatthewLundberg Interesting, thanks for that. Why was it deleted, sounds useful! That would explain why tests haven't caught it - I would probably have only thought to test NA with integers thinking that NA as double wouldn't make a difference. – Pachysandra

I can't answer that, but didn't know if you would try the merge with integers without the hint. The answerer claimed that character worked also, but I did not verify. – Caffey

@MatthewLundberg Yeah I needed the hint. Juggling too many balls currently. – Pachysandra

@MatthewDowle Just wanna say: I love how responsive you are to the community. Thanks for all the work you're doing! Keep it up. – Taliesin

Answer 2

Following up on comments in other answer, yes, here is the proof that it only affects type double columns (NA in integer and character columns are ok).

X = data.table(name=c("Raf","Jon","Ste","Rob","Smi","Joh"),
               depID=as.integer(c(31,33,33,34,34,NA)),key="depID")
Y = data.table(depID=as.integer(c(31,33,34,35)),
               depName=c("Sal","Eng","Cle","Mar"),key="depID")
Y[X]
   depID depName name
1:    NA      NA  Joh
2:    31     Sal  Raf
3:    33     Eng  Jon
4:    33     Eng  Ste
5:    34     Cle  Rob
6:    34     Cle  Smi

merge.data.frame(X,Y,all.x=T)
  depID name depName
1    31  Raf     Sal
2    33  Jon     Eng
3    33  Ste     Eng
4    34  Rob     Cle
5    34  Smi     Cle
6    NA  Joh    <NA>

Y = data.table(depID=as.character(c(31,33,34,35)),
               depName=c("Sal","Eng","Cle","Mar"),key="depID")
X = data.table(name=c("Raf","Jon","Ste","Rob","Smi","Joh"),
               depID=as.character(c(31,33,33,34,34,NA)),key="depID")
X
   name depID
1:  Raf    31
2:  Jon    33
3:  Ste    33
4:  Rob    34
5:  Smi    34
6:  Joh    NA
Y
   depID depName
1:    31     Sal
2:    33     Eng
3:    34     Cle
4:    35     Mar
str(X)
Classes ‘data.table’ and 'data.frame':  6 obs. of  2 variables:
 $ name : chr  "Raf" "Jon" "Ste" "Rob" ...
 $ depID: chr  "31" "33" "33" "34" ...
 - attr(*, "sorted")= chr "depID"
 - attr(*, ".internal.selfref")=<externalptr> 

merge.data.frame(X,Y,all.x=T)
  depID name depName
1    31  Raf     Sal
2    33  Jon     Eng
3    33  Ste     Eng
4    34  Rob     Cle
5    34  Smi     Cle
6  <NA>  Joh    <NA>

Y[X]
   depID depName name
1:    31     Sal  Raf
2:    33     Eng  Jon
3:    33     Eng  Ste
4:    34     Cle  Rob
5:    34     Cle  Smi
6:    NA      NA  Joh

---

THE PROBLEM HAS BEEN FIXED BY MATTHEW DOWLE IN V.1.8.7

Grizzled

Answer 3

Some info that can be usefull:

library(data.table);

X <- data.table(name=c("Raf","Jon","Ste","Rob","Smi","Joh"),depID=c(31,33,33,34,34,NA),key="depID")
#R) X
   #name depID
#1:  Joh    NA
#2:  Raf    31
#3:  Jon    33
#4:  Ste    33
#5:  Rob    34
#6:  Smi    34

Y <- data.table(depID=c(31,33,34,35),depName=c("Sal","Eng","Cle","Mar"),key="depID")
#R) Y
   #depID depName
#1:    31     Sal
#2:    33     Eng
#3:    34     Cle
#4:    35     Mar

#################
#LEFT OUTER JOIN#
#################
LJ <- merge.data.frame(X,Y,by="depID",all.x=TRUE); #by is implicit (see ?merge.data.frame)
#R) LJ
  #depID name depName
#1    31  Raf     Sal
#2    33  Jon     Eng
#3    33  Ste     Eng
#4    34  Rob     Cle
#5    34  Smi     Cle
#6    NA  Joh    <NA>

LJ2 <- Y[X];
#R) LJ2
   #depID depName name
#1:    NA      NA  Joh
#2:    31     Sal  Raf
#3:    33     Eng  Jon
#4:    33     Eng  Ste
#5:    34     Cle  Rob
#6:    34     Cle  Smi

##################
#RIGHT OUTER JOIN#
##################
RJ <- merge.data.frame(X,Y,by="depID",all.y=TRUE); #by is implicit (see ?merge.data.frame)
#R) RJ 
  #depID name depName
#1    31  Raf     Sal
#2    33  Jon     Eng
#3    33  Ste     Eng
#4    34  Rob     Cle
#5    34  Smi     Cle
#6    35 <NA>     Mar

RJ2 <- X[Y];
#R) RJ2
   #depID name depName
#1:    31  Raf     Sal
#2:    33  Jon     Eng
#3:    33  Ste     Eng
#4:    34  Rob     Cle
#5:    34  Smi     Cle
#6:    35   NA     Mar

#################
#FULL OUTER JOIN#
#################
FJ <- merge.data.frame(X,Y,all=T)
#R) FJ
  #depID name depName
#1    31  Raf     Sal
#2    33  Jon     Eng
#3    33  Ste     Eng
#4    34  Rob     Cle
#5    34  Smi     Cle
#6    35 <NA>     Mar
#7    NA  Joh    <NA>

FJ2 <- merge(X,Y,all=T)
#R) FJ2
   #depID name depName
#1:    NA  Joh      NA
#2:    31  Raf     Sal
#3:    33  Jon     Eng
#4:    33  Ste     Eng
#5:    34  Rob     Cle
#6:    34  Smi     Cle
#7:    35   NA     Mar

####################
#NATURAL INNER JOIN#
####################
IJ <- merge.data.frame(X,Y)
#R) IJ
  #depID name depName
#1    31  Raf     Sal
#2    33  Jon     Eng
#3    33  Ste     Eng
#4    34  Rob     Cle
#5    34  Smi     Cle

IJ2 <- merge(X,Y)
#R) IJ2
   #depID name depName
#1:    31  Raf     Sal
#2:    33  Jon     Eng
#3:    33  Ste     Eng
#4:    34  Rob     Cle
#5:    34  Smi     Cle


A <- data.table(time=as.POSIXct(c("10:01:01","10:01:02","10:01:04","10:01:05","10:01:02","10:01:01","10:01:01"),format="%H:%M:%S"),
                b=c("a","a","a","a","b","c","c"), 
                d=c(1,1.9,2,1.8,5,4.1,4.2));
B <- data.table(time=as.POSIXct(c("10:01:01","10:01:03","10:01:00","10:01:01"),format="%H:%M:%S"),b=c("a","a","c","d"), e=c(1L,2L,3L,4L));
setkey(A,b,time)
setkey(B,b,time)


###########
#ASOF JOIN#
###########
AOJ <- B[A,roll=T]
#R) AOJ
   #b                time  e   d
#1: a 2013-01-11 10:01:01  1 1.0
#2: a 2013-01-11 10:01:02  1 1.9
#3: a 2013-01-11 10:01:04  2 2.0
#4: a 2013-01-11 10:01:05  2 1.8
#5: b 2013-01-11 10:01:02 NA 5.0
#6: c 2013-01-11 10:01:01  3 4.1
#7: c 2013-01-11 10:01:01  3 4.2

Grizzled