Target gets duplicated depending on transformation input

[bart1]

Sorry for reporting all these strange edge cases, it took me a while to reduce the plan as much as possible from the large plan of the original analysis. It seems the bug is not reproduces if the targets br or dataTrainList are not included even though they seem independent. In this case depending on the input to the crossing of data the target dataTestList gets duplicated. It seems this duplication only occurs if the second element of cvo can be interpreted as a numeric (see the first two of the examples below). I expected in all cases below that only two targets for dataTestList would be generated.

require(drake)
#> Loading required package: drake
p<-drake_plan(trace=T,tidy_eval = T, transform = F,
  data = target(
    command = crossValOmit(radar, crossValOmission),
    transform = cross(radar=!!'dd',
      crossValOmission = !!cvo,
      .id = c(radar, crossValOmission)
    )
  ),
  br = target(
    command = annotate_model(data),
    transform = combine(data, 
                        .by = data)
  ),
  b = target(
    command = list(crossValId, data),
    transform = cross(data,
      crossValId = !!1,
      .id = c(radar, crossValOmission, crossValId)
    )
  ),
  a = target(
    command = list(b),
    transform = combine(b, .by = data)
  ),
  dataTrainList = target(
    command = list2(a, data),
    transform = map(a, data,
      .id = c(crossValOmission, radar)
    )
  ),
  dataTestList = target(
    command = list(a, data),
    transform = map(a, data,
      .id = c(crossValOmission, radar)
    )
  )
)
cvo<-c('a3', "2")
grep('dataTestList', transform_plan(p)$target, value=T)
#> [1] "dataTestList_2_dd"    "dataTestList_a3_dd"   "dataTestList_2_dd_2" 
#> [4] "dataTestList_a3_dd_2"
cvo<-c('3', "2")
grep('dataTestList', transform_plan(p)$target, value=T)
#> [1] "dataTestList_2_dd"   "dataTestList_3_dd"   "dataTestList_2_dd_2"
#> [4] "dataTestList_3_dd_2"
cvo<-c('3', "a2")
grep('dataTestList', transform_plan(p)$target, value=T)
#> [1] "dataTestList_3_dd"  "dataTestList_a2_dd"

^{Created on 2019-09-13 by the reprex package (v0.3.0)}

[wlandau]

Where is the variable dd?

[wlandau]

Oh, it's supposed to be just a string. Never mind...

[bart1]

yes sorry that was a quick hack to reduce the multiplication that was in the original plan

[wlandau]

The problem actually has to do with NA values. We need to remove them from the grid before we apply map(). Working on a patch, but a meeting is about to interrupt me...

[bart1]

No problem your amazingly responsive in any case.

[wlandau]

Should be fixed now. map() is now better at respecting the graph topology (e.g. nestings) of old grouping variables.