Error on trivial example

[DavorJ]

Hi Giulio, I am currently testing your package for consistency on some trivial cases, and got the following error with the below example with only 1 possible stratum per domain:

strata <- data.frame(
  "STRATO" = LETTERS[1:5],
  "N" = c(471114, 179296,  150211,  88104, 270184 + 20000)*100,
  "M1" = c(61.08, 89.19, 89.50, 76.41, 76.59),
  "S1" = c(19.91, 28.19, 35.03, 37.14, 31.40),
  "COST" = 1, "CENS" = 0, 
  "DOM1" = 1:5,
  "X1" = LETTERS[1:5]
)

solution <- SamplingStrata::optimizeStrata(
  errors = data.frame("DOM" = "DOM1", "CV1" = 0.05, "domainvalue" = 1:5),
  strata = strata, showPlot = FALSE, parallel = FALSE)


#> Error in solut[[3]] : subscript out of bounds

It seems to be related to this piece of code where if nrow(stcamp[[i]]) == 1 is true, the rbga object is removed from solut, but on line 208 it is referred to.

Also,, the errors dataframe needs a specific form, but this form is nowhere clearly documented as far as I can tell. Especially domainvalue is inconsistent. On one occasion it is used as reference to a column name as in buildFrameDF(), and sometimes as an indicator value as in this case. Maybe making a constructor function with dedicated help section for this error data.frame would make things easier?

[barcaroli]

Hi Davor, first of all, I would like to know if you are using the CRAN version or the GitHub one?

[DavorJ]

CRAN version.

[barcaroli]

You were right about the bug. I solved it by moving line 203 (rbga.object <- solut[[3]]) to line 193, before the check
if (nrow(stcamp[[i]]) == 1)
Now it should work, I obtain:
solution$aggr_strata
STRATO N M1 S1 COST CENS DOM1 SOLUZ
1 A 47111400 61.08 19.91 1 0 1 42.50148
2 B 17929600 89.19 28.19 1 0 2 39.95921
3 C 15021100 89.50 35.03 1 0 3 61.27628
4 D 8810400 76.41 37.14 1 0 4 94.50151
5 E 29018400 76.59 31.40 1 0 5 67.23171
It is a strange case, as in each domain, only one stratum is present, and there is no possible optimization (only application of the Bethel algorithm). Anyway, it is a possible case.
In the next days, I will submit this new bug-free version of SamplingStrata to CRAN.
About your comment regarding the difficulties in defining the errors dataframe, you are right, I will think about the development of a constructor function assisting the user. Thanks for the advice

[barcaroli]

The new version of SamplingStrata without the bug is now on CRAN

[barcaroli]

Hi Davor, I just added a function "buildErrorsDF" to the package in the GitHub version.
So, if you install from GitHub:

devtools::install_github("barcaroli/SamplingStrata")

then you can execute:

library(SamplingStrata)
strata <- data.frame(
"STRATO" = LETTERS[1:5],
"N" = c(471114, 179296, 150211, 88104, 270184 + 20000)*100,
"M1" = c(61.08, 89.19, 89.50, 76.41, 76.59),
"S1" = c(19.91, 28.19, 35.03, 37.14, 31.40),
"COST" = 1, "CENS" = 0,
"DOM1" = 1:5,
"X1" = LETTERS[1:5]
)
errors <- buildErrorsDF(strata = strata)
errors$CV1 <- rep(0.05,5)
checkInput(errors,strata)
solution <- SamplingStrata::optimizeStrata(
errors = errors,
strata = strata,
showPlot = FALSE,
parallel = FALSE)
solution$aggr_strata

[DavorJ]

There are a few other issues I encountered with custom strata and error dataframes:

• Depending on how the strata data.frame has been made, it could be that domains (DOM1 field) are not consecutive numbers which would result in an error due to this line:
  

  SamplingStrata/R/optimizeStrata.R

  Line 153 in 8bdb684

  for (i in (unique(strata$DOM1))) {

  
  Quick fix is to change it to: seq_along(unique(strata$DOM1)).

• Also in strata, the DOM1 field is assumed after M and S fields. If not, then these columns will be deleted here:
  

  SamplingStrata/R/optimizeStrata.R

  Lines 194 to 195 in 8bdb684

  	solut <- list(c(1), stcamp[[i]][c(1:grep("DOM1",
  	colnames(stcamp[[i]])))])

  
  Quick fix is: solut <- list(c(1), stcamp[[i]]) since it doesn't hurt for X fields to remain there?

• Another thing is that error dataframe assumes that domainvalue is the last column which may not be the case if it is constructed manually:
  

  SamplingStrata/R/optimizeStrata.R

  Line 156 in 8bdb684

  erro[[i]] <- erro[[i]][, -ncol(errors)]

  
  Safer is to remove by name: erro[[i]][["DOMAINVALUE"]] <- NULL

I don't know whether these proposals would break anything else on other places. It wouldn't hurt to have some automatic test cases to test against various scenario's, e.g. with testthat.

I switched to the GitHub version and fixing issues on the go as I encounter them. Depending on later needs, I may fork it, or if you prefer to keep things here feel free to make me a dedicated branch (e.g. davor) where I can make changes without impacting your main branch.

[barcaroli]

I just created a new branch ("davor")