CODEWITHSUNDEEP
rdata.table
andrew
andrew

Reputation: 2574

Using by with data.table

Here I am trying to use the by parameter in the data.table to rank the prediction column within each group. I haven't been able to understand why the following piece of code isn't working:

> x.small
       prediction group
 1: -0.0093753015    up
 2:  0.0204832283  down
 3: -0.0091790179  down
 4: -0.0473988803  down
 5:  0.0144955868  down
 6: -0.0139455871  down
 7:  0.0005746896    up
 8: -0.0174406693  down
 9: -0.0180556244  down
10: -0.0343069464    up
> x.small[, rank(prediction), by=group]
Error in rank(prediction) :
  'names' attribute [7] must be the same length as the vector [3]

But this example code works fine:

> diamonds.dt <- data.table(diamonds[1:10, c('carat', 'color')])
> diamonds.dt
    carat color
 1:  0.23     E
 2:  0.21     E
 3:  0.23     E
 4:  0.29     I
 5:  0.31     J
 6:  0.24     J
 7:  0.24     I
 8:  0.26     H
 9:  0.22     E
10:  0.23     H
> diamonds.dt[, rank(carat), by=color]
    color  V1
 1:     E 3.5
 2:     E 1.0
 3:     E 3.5
 4:     E 2.0
 5:     I 2.0
 6:     I 1.0
 7:     J 2.0
 8:     J 1.0
 9:     H 2.0
10:     H 1.0

Any help would be much appreciated!

EDIT:

Okay now I really have no idea what's going on, this is very bizarre. I tried making a reproducible example for @Ananda but could not recreate the error. I even tried running the ranking logic on an exact copy of the prediction column and it worked fine:

> x.small[, prediction.copy:=prediction]
> x.small[, rank(prediction.copy), by=group]
    group V1
 1:    up  2
 2:    up  3
 3:    up  1
 4:  down  7
 5:  down  5
 6:  down  1
 7:  down  6
 8:  down  4
 9:  down  3
10:  down  2
> x.small[, rank(prediction), by=group]
Error in rank(prediction) :
  'names' attribute [7] must be the same length as the vector [3]

How could there be two different results from two identical columns?

EDIT 2:

Output of dput(x.small):

> dput(x.small)
structure(list(prediction = structure(c(-0.00937530151309606,
0.0204832283018108, -0.00917901792827827, -0.0473988802836657,
0.0144955868466372, -0.0139455871394683, 0.000574689607249577,
-0.0174406692627376, -0.0180556244204637, -0.0343069463869563
), .Names = c("1", "2", "3", "4", "5", "6", "7", "8", "9", "10"
)), group = c("up", "down", "down", "down", "down", "down", "up",
"down", "down", "up"), prediction.copy = c(-0.00937530151309606,
0.0204832283018108, -0.00917901792827827, -0.0473988802836657,
0.0144955868466372, -0.0139455871394683, 0.000574689607249577,
-0.0174406692627376, -0.0180556244204637, -0.0343069463869563
)), .Names = c("prediction", "group", "prediction.copy"), row.names = c(NA,
-10L), class = c("data.table", "data.frame"), .internal.selfref = <pointer: 0x22f2af8>)

Upvotes: 0

Views: 179

Answers (1)

andrew
andrew

Reputation: 2574

I guess I'll just close this one. If you are having this same issue, check if the problem column is a named vector by running str(x.small) and seeing if the vector starts with the word "Named". For some reason using the by parameter when operating on a named vector is causing issues. This appears to be a minor bug in earlier versions of data.table that was patched in later versions. To fix it, upgrade data.table or just use unname() as @Frank suggested:

x.small[,rank(unname(prediction)), by=group]

Upvotes: 1

Related Questions