Long to Wide issue

[Pmassey]

I am running into an issue when using the apollo_longToWide in version 0.3.9.

When I run it I get the following error:

Error in rle(generic_pattern$ID_obs: ‘x’ must be a vector of atomic type

In addition:
In ny*ax : NAs produced by integer overflow

Any help is appreciated!

[stephanehess]

Hi

can you provide us with more information, such as the column names and data types in the original data, plus the call you make with the settings, etc

Thanks

Stephane

[Pmassey]

Yes, sorry! I cannot include any data, as they are held in a secure server. These are panel data at the individual, year level.

id_num: (int) identifies the indiviudal across years
choice_id: (int) is a unique id_num-year grouping
alt_num: (int) is the alternatives in each choice set
choice: (True/False) is TRUE for the choice made and FALSE otherwise.
List of ASC's: I have a list of ASC's that are all integer values.

The call I use is:

Code: Select all

database <- apollo_longToWide(longData = as.data.frame(choice_sets), longToWide_settings = list(
    idColumn = "id_num", obsColumn = "choice_id", altColumn = "alt_num",
    choiceColumn = "choice",
    altSpecAtts = list of ASC's))

The weird part is this was all working when I was using version 0.3.6, however I recently got the secure server to update the version to the latest available and now the issue has come up.

[jackhan1208]

Hi both,
I encountered the same issues when using the latest Apollo package 0.3.9
I had no such issues in earlier version.
Since I couldnt figure the reason why when I have identical dataset but able to run in earlier version , but not the latest version.
I looked into the source code of the longtowide function
and found one particular issue in the source code.

Code: Select all

longData$ID_obs <- interaction(
  longData[, idColumn],
  longData[, obsColumn],
  drop = TRUE,
  lex.order = TRUE
)

In the source code , we can see that apollo produce internal ID_obs as factor variable by default when using interaction () function.
However, the subsequent code

Code: Select all

rle(generic_part$ID_obs)

rle () function does not directly accept "factor" format variable, therefore, I encountered the same issues.

rle(generic_pattern$ID_obs: ‘x’ must be a vector of atomic type

A workaround that worked in my case is to make a local copy of apollo function and an apollo environment, then add rle(as.character(x) to allow rle function to accept factor variables. basically, a override of the function without change the source code, of course, this can be changed in source code I believe.

Code: Select all

# Create a local copy of Apollo's function
apollo_longToWide_fixed <- apollo::apollo_longToWide

# Create an environment inheriting Apollo's namespace
fix_env <- new.env(
  parent = environment(apollo_longToWide_fixed)
)

# Override rle() locally to handle factor variables
fix_env$rle <- function(x) {
  base::rle(as.character(x))
}

# Apply the fix only to the copied function
environment(apollo_longToWide_fixed) <- fix_env

I hope this helps

Kind regards
Jack

[stephanehess]

Hi

this is indeed a bug in 0.3.9 which is being fixed for 0.4.0. A development version should be on the website within a week or so, and the official release in early November

Stephane