Improvements

• Make work with all versions of xgboost. (#267)

Improvements

• Addsstep_lencode() to perform analytical likelihood encoding. (#258)

• Addssmooth argument tostep_lencode() to allow for partial pooling in numeric outcomes. (#261)

• step_feature_hash() has been fully deprecated in favor oftextrecipes::step_dummy_hash(). (#253)

Improvements

• step_umap() has tunableinitial andtarget_weight arguments. (#223,#222)

• All messages, warnings and errors has been translated to use {cli} package (#153,#155).

Improvements

• step_umap() has gainedinitial andtarget_weight arguments. (#213)

• Calling?tidy.step_*() now sends you to the documentation forstep_*() where the outcome is documented. (#216)

• Documentation for tidy methods for all steps has been improved to describe the return value more accurately. (#217)

• {keras} and {tensorflow} have been moved to Suggests instead of Imports. (#218)

• step_collapse_stringdist() will now return predictors as factors. (#204)

• Fixed regression from 1.1.2 instep_lencode_glm() where it couldn't be used on multiple columns.

Improvements

• Thekeep_original_cols argument has been added tostep_woe(). This change should mean that every step that produces new columns has thekeep_original_cols argument. (#194)

• Many internal changes to improve consistency and slight speed increases.

Breaking Changes

• step_pca_sparse(),step_pca_truncated() andstep_pca_sparse_bayes() now returns data unaltered ifnum_comp = 0. This is done to be consistent with recipes steps of the same nature. (#190)

Bug Fixes

• Fixed bug wherestep_pca_truncated() didn't work with zero selection. (#181)

• The tidy() methods forstep_discretize_cart(),step_discretize_xgb(),step_embed(),step_feature_hash(),step_lencode_bayes(),step_lencode_glm(),step_lencode_mixed(),step_pca_sparse(),step_pca_sparse_bayes(),step_pca_truncated(),step_umap(), andstep_woe() now correctly return zero-row tibbles when used with empty selections. (#181)

New Steps

• step_pca_truncated() has been added. This step only calculates the components that are required, and will be a speedup in cases where it is used on many variables. (#82)

Improvements

• step_collapse_stringdist() has gainedmethod andoptions arguments to allow for different types of string distance calculations. (#152)

• step_umap() has gained the argumentmetric. (#154)

• step_embed() has gained thekeep_original_cols argument. (#176)

• All steps now haverequired_pkgs() methods.

• Steps with tunable arguments now have those arguments listed in the documentation.

• All steps that add new columns will now informatively error if name collision occurs.

• step_collapse_cart() can pool a predictor's factor levels using a tree-based method.

• step_collapse_stringdist() can pool a predictor's factor levels using string distances.

• Case weights support have been added tostep_discretize_cart(),step_discretize_xgb(),step_lencode_bayes(),step_lencode_glm(), andstep_lencode_mixed().

• step_embed() now correctly defaults to have a random id with the word "embed". (#102)

• step_feature_hash() is soft deprecated in embed in favor ofstep_dummy_hash() in textrecipes. (#95)

• Steps now have a dedicated subsection detailing what happens whentidy() is applied. (#105)

• Reorganize documentation for all recipe steptidy methods (#115).

• Fixed a bug wherewoe_table() andstep_woe() didn't respect the factor levels of the outcome. (109)