superbData fails when called without loading the package with superb::ERROR(22)

[achetverikov]

superbData looks for CI.mean in the global environment, fails when package is not loaded or only namespace-loaded with requireNamespace

From a fresh R session:

> data(ToothGrowth)
+ superb::superbData(ToothGrowth, BSFactors = c("dose", "supp"), 
+   variables = "len")
superb::FYI: The variables will be plotted in that order: dose, supp (use factorOrder to change).
Error in `superbPlot()`:
! superb::ERROR(22): The function CI.mean is not a known function for error bars. Exiting...

# still fails
> requireNamespace('superb')
+ superb::superbData(ToothGrowth, BSFactors = c("dose", "supp"), 
+   variables = "len")
Loading required namespace: superb
superb::FYI: The variables will be plotted in that order: dose, supp (use factorOrder to change).
Error in `superbPlot()`:
! superb::ERROR(22): The function CI.mean is not a known function for error bars. Exiting...

# after loading with library, the example works
> library(superb)
Warning message:
package ‘superb’ was built under R version 4.5.2 
> data(ToothGrowth)
+ superb::superbData(ToothGrowth, BSFactors = c("dose", "supp"), 
+   variables = "len")
superb::FYI: The variables will be plotted in that order: dose, supp (use factorOrder to change).
$summaryStatistics
  dose supp center lowerwidth upperwidth
1  0.5   OJ  13.23  -3.190283   3.190283
2  0.5   VC   7.98  -1.964824   1.964824
3    1   OJ  22.70  -2.797727   2.797727
4    1   VC  16.77  -1.799343   1.799343
5    2   OJ  26.06  -1.899314   1.899314
6    2   VC  26.14  -3.432090   3.432090
...

Environment:

• superb 1.0.1
• R 4.5.1 (Windows)

[VIC-Laboratory-ExperimentalData]

I added an experimental feature one year ago (superb version 0.95.23 and above) whereby it is possible to give a namespace explicitly when the namespace of the error bar function is in an unexpected Namespace.

data(ToothGrowth)

superb::superbData(ToothGrowth, 
    BSFactors = c("dose", "supp"), 
    variables = "len",
    statistic = "superb::mean" ## experimental: give an explicit namespace for the CI
)

This experimental feature was not extensively tested so I you see issues with it, let me know.

[achetverikov]

super, thanks, will test this!

[achetverikov]

This works but not always:

> data(sleep, package = "datasets")
> sleep_wide <- reshape(sleep, direction = "wide", idvar = "ID", timevar = "group")

> superb::superbData(
  sleep_wide,
  WSFactors   = "group(2)",
  variables   = c("extra.1", "extra.2"),
  adjustments = list(purpose = "single", decorrelation = "CM"),
  statistic   = "superb::mean",
  errorbar    = "superb::CI"
)

superb::EXPERIMENTAL(1): statistic given with namespace superb::mean
superb::EXPERIMENTAL(2): errorbar given with namespace superb::CI
Error in `get()`:
! object 'HyunhFeldtEpsilon' of mode 'function' was not found

And this seem to create superbERRORBAR and superbSTATISTIC in the global environment is undesirable more generally (see CRAN policies: "Packages should not modify the global environment (user’s workspace). " https://cran.r-project.org/web/packages/policies.html or the CRAN cookbook https://contributor.r-project.org/cran-cookbook/code_issues.html#writing-to-the-.globalenv)

[achetverikov]

see proposed PR #10

[VIC-Laboratory-ExperimentalData]

Thanks for looking into it and proposing a solution! I'll double check everything early next month.

[achetverikov]

Thanks! If you then push the updates to CRAN, that would be great, I want to submit a small package there that uses superb for a bit of processing and currently assumes this is fixed.

[achetverikov]

Hi @VIC-Laboratory-ExperimentalData ! Were you able to check the PR #10 ?