V 1.0.5, July 2022

FEAT: - New functions compute_probability_ratio and compute_weight_of_evidence to be used for target encoding - New function get_most_frequent_element to identify most frequent element in a list

V 1.0.4

BUGFIX: Fix generate_from_character, when there were some NAs in the column it would drop the line. It is not the case anymore.

V 1.0.3

BUGFIX: Fix bud on fast_is_bijection when column has multiple class FEAT: Harmonize logging levels between functions

V 1.0.2

Remove useless dependencies. Make sure library works on windows, macos, ubuntu, and R versions from 3.3 to 4.1.

V 1.0.1

Based on CRAN feedbacks removed problematic vignettes.

V 1.0.0

For this version 1.0.0 there are a lot of changes, and version is not compatible with previous version of the paclage.

Also there might be some rework to do on code using previous version of this package (and we are orry about it), we strongly believe that this version will be easier to use, faster, and more maintanable in time.

In this version : - All function names and variables are snake_case (there used to be a mix of camel case and snake case) - We remove a lost of useless code that was slowing done the package (particularly garbage collection) - We made the code more readable so that it is easier to contribute to this package - Logging is more explicit and cleaner. - We took into account linting. - A few more functions are availables.

We hope that you will like even more this new version of the package. Please don’t hesitate to provide feedback, warn us about bug, suggest improvements or even better developp some improvements on this package. To do so please go to github (https://github.com/ELToulemonde/dataPreparation/).

V 0.4.3

• Fix :
  • In same_shape: there was a future bug due to change in class “matrix”. Fixed it by implementing 2 functions to check class

V 0.4.2

• Fix test :
  • Case in build_encoding: min_frequency allows to drop rare values” was not built correctly.

V 0.4.1

• New features:
  • New functions:
    • Functions target_encode and build_target_encoding have been implemented to provide target encoding which is the process of replacing a categorical value with the aggregation of the target variable.
    • Function remove_sd_outlier helps to remove rows that have numerical values to extrem.
    • Function remove_percentile_outlier helps to remove rows that have numerical values to extrem (based on percentile analysis).
    • Function remove_rare_categorical helps to remove rows that have categorial values to rare.
  • New features in existing functions :
    • Function prepare_set integrate target_encode function. It is called by providing target_col and target_encoding_functions.

V 0.4.0

• New features:
  • New features in existing functions :
    • To avoid issues based on column names, we will check and rename columns that have same names.
    • In aggregate_by_key generated column names are changed to be more explicit.
    • In aggregate_by_key generated from character column with more than values is now count of unique instead of count.
    • Added missing auto default values on cols
• Bug fixes:
  • which_are_bijection and which_are_in_double are using bi_col_test which was not working with 2 column data set. It is fixed.
  • prepare_set optinal argumennt factor_date_type was not working. It is fixed.
• Other changes:
  • Changed which_are_included example since it was to slow for CRAN. Also it might be a little bit more explicit now.
  • Changed aggregate_by_key example since it was to slow for CRAN.
• Integration:
  • Rewrite all tests to make them more readable
  • Code coverage is improved, depencies on messy_adult set is lowered

WARNING: - In aggregate_by_key generated column names are changed. - In aggregate_by_key generated column for character is different.

V 0.3.9

• Integration:
  • Matching new devtools requierments
  • Starting to rewrite unittest to make it more readable

V 0.3.8

• New features:
  • New features in existing functions:
    • Identification of bijection through internal function fast_is_bijection is way faster (up to 40 times faster in case of bijection). So whichArebijection and fastFiltervariables are also improved.
    • Remove remaining gc to save time.
    • In one_hot_encoder added parameter type to make choise between logical or numerical results.

V 0.3.7

• New features:
  • New functions:
    • Function as.POSIXct_fast is now available. It helps to transform to POSIXct way faster (if the same date value is present multiple times in the column).
  • New features in existing functions:
    • In dates identifications, we make it faster by computing search of format only on unique values.
    • In date transformation, we made it faster by using as.POSIXct_fast when it is necessary.
    • Functions findAndTransFormDates, find_and_transform_numerics and un_factor now accept argument cols to limitate search.
• Bug fixes:
  • Control that over-allocate option is activated on every data.table to avoid issues with set. Package should be more robust.
  • In bijection search (internal function fast_is_bijection) there was a bug on some rare cases. Fixed but slower.

-Code quality: - Improving code quality using lintr - Suppressing some useless code - Meeting new covr standard - Improve log of setColAsXXX

V 0.3.6

• Bug fixes:
  • identify_dates had a weird bug. Solved
• Integration:
  • Making dataPreparation compatible with testthat 2.0.0

V 0.3.5

• New features:
  • New features in existing functions:
    • findAndTransFormDates now as an ambiguities parameter, IGNORE to work as before, WARN to check for ambiguities and print them, SOLVE to try to solve ambiguities on more lines.
    • one_hot_encoder now uses a build_encoding functions to be able to build same encoding on train and on test.
    • aggregate_by_key is now way faster on numerics. But it changed the way it gets input functions.
    • fast_scale now as a way parameter which allow you to either scale or unscale. Unscaling numeric values can be very usefull for most post-model analysis.
    • set_col_as_date now accept multiple formats in a single call.
  • New functions:
    • build_encoding build a list of encoding to be used by one_hot_encoder, it also has a parameter min_frequency to control that rare values doesn’t result in new columns.
    • Previously private function identify_dates is now exported. To be able to perform same transformation on train and on test.
    • Adding dataPreparationNews function to open NEWS file (inspired from rfNews() of randomForest package)
• Bug fixes:
  • findAndTransFormDates: bug fixed: user formats weren’t used.
  • identify_dates: some formats where tested but would never work. They have been removed.
• Refactoring:
  • Unit test partly reviewed to be more readable and more efficient. Unit test time as been divided by 3.
  • Improving input control for more robust functions

WARNING: - one_hot_encoder now requires you to run build_encoding first. - aggregate_by_key now require functions to be passed by character name

This version is making (as much as possible) transformation reproducible on train and test set. This is to prepare future pipeline feature.

V 0.3.4

• Improvement of function
  • which_are_bijection: It is 2 to 15 time faster than previous version.
  • which_are_included: It is a bit faster.
• Bug fixes:
  • generate_factor_from_date: default value was missing. Fixed.
• New features:
  • New features in existing functions:
    • fast_filter_variables has a new parameter (level) to choose which types of filtering to perform

WARNING: - which_are_included: in case of bijection (col1 is a bijection of col2), they are both included in the other, but the choice of the one to drop might have changed in this version.

V 0.3.3

• New features:
  • New features in existing functions:
    • findAndTransFormDates now recognize date character even if there are multiple separator in date (ex: “2016, Jan-26”).
    • findAndTransFormDates now recognize date character even if there are leading and tailing white spaces.

WARNING: - date3 column in messy_adult data set has changed in order to illustrate the recognition of date character even if there are leading and/or trailing white spaces. - date4 column in messy_adult data set has changed in order to illustrate the recognition of date character even if there are multiple separator.

V 0.3.2

• Change URLs to meet CRAN requirement

v 0.3.1

• Fix bug in Latex documentation

v 0.3

• New features:
  • New features in existing functions:
    • findAndTransFormDates now recognize date character even if “0” are not present in month or day part and month as lower strings.
    • findAndTransFormDates and set_col_as_date now work with factors.
  • New functions:
    • fast_discretization: to perform equal freq or equal width discretization on a data set using data.table power.
    • fast_scale: to perform scaling on a data set using data.table power.
    • one_hot_encoder: to perform one_hot encoding on a data set using data.table power.
  • New documentation:
    • A new vignette to illustrate how to build a correct train and test set unising data preparation
• Minor changes in log (in particular regarding progress bars and typos)
  • Due to dependencies issues with tcltk, we stop using it and start using progress
• Refactoring:
  • Private function real_cols take more importance to control that columns have the correct types and handling “auto” value.
  • Making code faster: some functions are up to 30% faster
  • Review unit testing to be faster
  • Unit test evolution to be more readable

WARNING: - date1 column in messy_adult data set has changed in order to illustrate the recognition of date character even if “0” are not present in month or day part.

v 0.2

• Improving unit testing and code coverage
• Improving documentation
• Solving minor bug in date conversion and in which functions
• New features:
  • New functions:
    • un_factor to unfactor columns, when reading wasn’t performed in expected way.
    • same_shape to make ure that train and test set have exactly the same shape.
    • generate new columns from existing columns (generate functions)
      • generate factor from dates: generate_factor_from_date
      • diffDates becomes generate_date_diffs (for better name understanding).
      • generate numerics and booleans from character of fators (using generate_from_factor and generate_from_character)
    • set_col_as_factor a function to make multiple columns as factor and controling number of unique elements
  • New features in existing functions:
    • which functions: add keep_cols argument to make sure that they are not dropped
    • fast_filter_variables: verbose can be T/F or 0, 1, 2 in order to control level of verbosity
    • findAndTransFormDates and set_col_as_dates now recognize and accept timestamp.

WARNING: - If you were using diffDates, it is now called generate_date_diffs - date2 column in messy_adult data set have changed in order to illustrate new timestamp features - set_col_as_factorOrLogical doesn’t exist anymore: it as been splitted between set_col_as_factor and generateFromCat - Considering all those changes: shape_set and prepare_set don’t give the same result anymore.

v 0.1: release on CRAN July 2017