Changelog

wranglr 0.0.3

Breaking changes

• create_sumry_tbl() now returns Q25 and Q75 columns in place of the IQR column. The quartiles do not assume a symmetric distribution. To get the old value, use Q75 - Q25.

• create_sumry_tbl() now returns the count of non-missing values in n, and the count of all values in the new total_n column. n and NAs now sum to total_n. If you used n as the total count, use total_n.

• imputeNAs() is now impute_median(). Its behaviour does not change. Replace all calls to imputeNAs() with impute_median().

• impute_predictors() is now defunct and gives an error when you call it, it will be fully deprecated/removed in future versions.

New features

• New impute_missing() replaces NA values in a numeric vector with random draws from a fitted distribution. It estimates the distribution parameters with maximum likelihood. distr selects a Gaussian ("norm", the default), Gamma ("gamma"), or Beta ("beta") distribution, and seed makes the draws reproducible.

  x <- rnorm(10, mean = 10, sd = 5)
  x[c(2, 7)] <- NA_real_
  impute_missing(x)                   # Gaussian
  impute_missing(abs(x), "gamma")     # Gamma

Minor improvements and fixes

• create_sumry_tbl() now returns Q25 and Q75 columns in place of IQR, and a new total_n column. n now counts only non-missing values.

• create_sumry_tbl() now keeps the class and level order of the first grouping column if it is a factor. The “Total” row is the last level. Other column types become character, so numeric grouping columns, such as cyl in mtcars, now work without conversion to a factor.

• create_sumry_tbl() now returns NA for all statistics (except total_n, NAs, and n) when all values of var are missing, instead of an error.

wranglr 0.0.2

New

• Updated and improved create_summ_tbl()
  • new name!
    • now is create_sumry_tbl()
  • now uses dplyr::group_by() under the hood for ease of use
  • new snapshot tests
• New methods for imputeNAs()
  • character and factor S3 methods
  • imputes randomly by sampling according to the distribution of unique values in x and replaces
  • now the .create_strata() function can use imputeNAs()
  • does so internally using internal methods rather than on-the-fly
• S3 methods for .get_indices()
  • now dispatches correctly using S3 methods rather than switch() and complex logic

Bugs

• Fixed major bug in imputeNAs()
  • data frame method was indexing incorrectly
  • removed seq() call and index only on p
  • this was a bad one …

Improvements

• Create vfold_splits() unit tests and minor syntax clean up
  • added new unit tests (snapshots) for vfold_splits()
  • fixed a bug that was caused by using the names of breaks after An unlist
    • now uses a tmp strat_vars variable
  • all snapshots included under the vfold_splits variant
• Simplified .get_indices.default()
  • no just returns if unknown class
  • no more NULL for no stratification
  • this is now handled upstream, in .vfold_splits()
• Removed the idx param from .get_indices() methods
  • this was superfluous and a little hacky
  • solution isn’t great but considering it is just in the single data frame use case, maybe acceptable
• Cleaned up create_kfold()
  • updated unit tests and simplified code
  • streamline methods, especially character and factor for .get_indices()
  • breaks now cannot be NA

Removed

• no longer imports the purrr package:
  • no longer uses purrr::transpose()
  • uses base R and Map() instead to invert lists
  • now uses helpr::piter() over purrr::pmap()
• removed all mention of term “adat” and nuke S4 methods
  • the 'soma_adat' class is also a data frame so this shouldn’t affect it too much to remove the method explicitly. Should cascade into the date frame method
  • checks in place to ensure attributes are not stripped

wranglr 0.0.1 🎉

• Initial release! 🥳
  • this is v0.0.1 tagged and released