Example for step_novel() now better illustrates how it works. (@Edgar-Zamora, #1248)
recipe(), prep(), and bake() now work with sparse tibbles. (#1364, #1366)
recipe(), prep(), and bake() now work with sparse matrices. (#1364, #1368, #1369)
prep.recipe(..., strings_as_factors = TRUE) now only converts string variables that have role "predictor" or "outcome". (@dajmcdon, #1358, #1376)
Improved error message for misspelled argument in step functions. (#1318)
recipe() can now take data.frames with list-columns or sf data.frames as input to data. (#1283)
recipe() will now show better error when columns are misspelled in formula (#1283).
add_role() now errors if a column would simultaneously have roles "outcome" and "predictor". (#935)
prep() will now error if the ptype of the data doesn't match which was used to define the recipe. (#793)
Added more documentation in ?selections about how tidyselect::everything() works in recipes. (#1259)
New extract_fit_time() method has been added that returns the time it took to train the recipe. (#1071)
step_spline_b(), step_spline_convex(), step_spline_monotone(), and step_spline_nonnegative() now throws informative errors if thedegree, deg_free, and complete_set arguments causes an error. (#1170)
step_mutate() gained .pkgs argument to specify what packages need to be loaded for step to work. (#1282)
step_interact() now gives better error if terms isn't a formula. (#1299)
The prefix argument of step_dummy_multi_choice() is now properly documented. (#1298)
Significant speedup in step_dummy() when applied to many columns. (#1305)
step_dummy() now gives an informative error on attempt to generate too many columns to fit in memory. (#828)
step_dummy() and step_unknown() now throw more informative warnings for unseen levels. (#450)
step_dummy() now throws more informative warnings for NA values. (#450)
step_date() now accepts "mday" as a possible feature. (@Edgar-Zamora, #1211)
NA levels in factors aren't dropped when passed to recipe(). (#1291)
recipe() no longer crashes when given long formula expression (#1283).
Fixed bug in step_ns() and step_bs() where knots field in options argument wasn't correctly used. (#1297)
Bug fixed in step_interact() where long formulas were used. (#1231, #1289)
Fixed documentation mistake where default value of keep_original_cols argument were wrong. (#1314)
Developer helper function recipes_ptype() has been added, returning expected input data for prep() and bake() for a given recipe object. (#1329)
Developer helper function recipes_ptype_validate() has been added, to validate new data is compatible with recipe ptype. (#793)
Developer helper functions recipes_names_predictors() and recipes_names_outcomes() have been added to aid variable selection in steps. (#1026)
step_log() breaks legacy recipe objects by indexing names(object) in bake(). (@stufield, #1284)Minor speed-up and reduced memory consumption for step_pca() in the bake() stage by reducing unused multiplications (@jkennel, #1265)
Document that update_role(), add_role() and remove_role() are applied before steps and checks. (#778)
Documentation for tidy methods for all steps has been added when missing and improved to describe the return value more accurately. (#936)
step_dummy() will now error if passed character instead of loudly ignoring them. Only applicable when setting strings_as_factors = FALSE. (#1233)
It is now documented that step_spline_b() can be made periodic. (#1223)
prep() now correctly throws a warning when training argument is set when prepping a prepped recipe, telling the user that it will be ignored. (#1244)
When errors are thrown about wrongly typed input to steps, the offending variables and their types are now listed. (#1217)
All warnings and errors have been updated to use the cli package for increased clarity and consistency. (#1237)
Added warnings when step_scale(), step_normalise(), step_center() or step_range() result in NaN columns. (@mastoffel, #1221)
Fixed bug where step_factor2string() if strings_as_factors = TRUE is set in prep(). (#317)
Fixed bug where tidy.step_cut() always returned zero row tibbles for trained recipes. (#1229)
spline2_apply (#1200)step_ns(), step_bs(), step_spline_b(), step_spline_convex(), step_spline_monotone(), step_spline_natural(), step_spline_nonnegative()) would error if baked with 1 row. (#1191)step_classdist_shrunken(), a regularized version of step_classdist(), was added. (#1185)step_bs() and step_ns() have gained keep_original_cols argument. (#1164)
The keep_original_cols argument has been added to step_classdist(), step_count(), step_depth(), step_geodist(), step_indicate_na(), step_interact(), step_lag(), step_poly(), step_regex(), step_window(). The default for each step is set to preserve past behavior. This change should mean that every step that produces new columns has the keep_original_cols argument. (#1167)
Fixed bugs where step_classdist(), step_count(), step_depth(), step_geodist(), step_interact(), step_nnmf_sparse(), and step_regex() didn't work with empty selection. All steps now leave data unmodified when having empty selections. (#1142)
step_classdist(), step_count() and step_depth() no longer returns a column with all NAs with empty selections. (#1142)
step_regex() no longer returns a column with all 0s with empty selections. (#1142)
The tidy() methods for step_geodist(), step_nnmf_sparse(), and step_sample() now correctly return zero-row tibbles when used with empty selections. (#1144)
step_poly_bernstein(), step_profile(), step_spline_b(), step_spline_convex(), step_spline_monotone(), step_spline_natural(), and step_spline_nonnegative() now correctly return a zero row tibble when used with empty selection. (#1133)
Fixed bug where the tidy() method for step_sample() didn't return an id column. (#1144)
check_class(), check_missing(), check_new_values(), check_range(), step_naomit(), step_poly_bernstein(), step_spline_b(), step_spline_convex(), step_spline_monotone(), step_spline_natural(), step_spline_nonnegative(), and step_string2factor() now throw an informative error if needed non-standard role columns are missing during bake(). (#1145)
step_window() now throws an error instead of silently overwriting if names argument overlaps with existing columns. (#1172)
step_regex() and step_count() will now informatively error if name collision occurs. (#1169)
Added developer function remove_original_cols() to help remove original columns that are no longer needed. (#1149)
Added developer function recipes_remove_cols() to provide standardized way to remove columns by column names. (#1155)
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. (#983)
Fixed bug in step_spline_b(), step_spline_convex(), step_spline_monotone(), and spline_nonnegative() where you weren't able to tune the degree argument.
step_range() now perform correctly performs clipping on recipes created before 1.0.3. (#1097)
tidy() method for step_impute_mean(), step_impute_median(), and step_impute_mode() now the imputed value with the column name value instead of model. This is in line with the output of step_impute_lower(). (#826)Added outside argument to step_percentile() to determine different ways of handling values outside the range of the training data.
step_range() is now backwards compatible with respect to the clipping argument that was added 1.0.3, and old saved recipes can now be baked. (#1090)
update print methods to use cli package for formatting. (#426)
Print methods no longer errors for untrained recipes with long selections. (#1083)
The recipe, step, and check methods for generics::tune_args() are now registered unconditionally (tidymodels/workflows#192).
Added a conditionMessage() method for recipes_errors to consistently point out which step errors occurred in when reporting errors. (#1080)
Added missing tidy method for step_intercept() and step_lag(). (#730)
Errors in prep() and bake() will now indicate which step caused the error. (#420)
Developer focused check_type() got a new types argument for more precise checking of column types.
recipes_extension_check() have been added. This developer focused function checks that steps have all the required S3 methods.
recipe() now error more informatively when data is missing. (#1042)
step_dummy() no longer returns integer columns as there are a number of contrast methods that return fractional values. (#1053)
Fixed a 0-length recycling bug in step_dummy_extract() exposed by the
development version of purrr (#1052).
Types of variables have been made granular. "nominal" has been split into "ordered" and "unordered" and "numeric" has been split into "double" and "integer". (#993)
New selectors: all_double(), all_ordered(), all_unordered(), all_date() and all_datetime(), in addition to the existing all_numeric() and all_nominal(). All selectors come with a *_predictors() variant. (#993)
Developer focused .get_data_types() generic has been added to designate types of columns. Exported for use in extension packages that deal with types not supported in recipes directly. (#993)
The step_date() function now defaults to using the clock package to format day-of-week and month labels. (#1048)
step_range() has gained a argument clipping that when set to FALSE no longer clips the data to be between min and max.
Added documentation regarding developer functions ?developer_functions. (#1163)
A new set of basis functions were added: step_spline_b(), step_spline_convex(), step_spline_monotone(), step_spline_natural(), step_spline_nonnegative(), and
step_poly_bernstein().
step_date(), step_dummy(), step_dummy_extract(), step_holiday(), step_ordinalscore(), and step_regex() now returns integer results when appropriate. (#766)
The default for the strict argument in step_integer() has been changed from FALSE to TRUE. The function will thus return integers, rather than whole-number numerics, by default. (#766)
The default for the value argument in step_intercept() has been changed from 1 to 1L. (#766)
step_holiday() didn't work if it isn't have any missing values. (#1019)Added support for case weights in the following steps
step_center()step_classdist()step_corr()step_dummy_extract()step_filter_missing()step_impute_linear()step_impute_mean()step_impute_median()step_impute_mode()step_normalize()step_nzv()step_other()step_percentile()step_pca()step_sample()step_scale()A number of developer focused functions to deal with case weights are added: are_weights_used(), get_case_weights(), averages(), medians(), variances(), correlations(), covariances(), and pca_wts()
recipes now checks that all columns in the data supplied to recipe() are also present in the new_data supplied to bake(). An exception is made for columns with roles of either "outcome" or "case_weights", which are typically not required at bake() time. The new update_role_requirements() function can be used to adjust whether or not columns of a particular role are required at bake() time if you need to opt out of this check (#1011).
The summary() method for recipe objects now contains an extra column to indicate which columns are required when bake() is used.
step_time() has been added that extracts time features such as hour, minute, or second. (#968)Fixed bug in which functions that step_hyperbolic() uses (#932).
step_dummy_multi_choice() now respects factor-levels of the selected variables when creating dummies. (#916)
step_dummy() no works correctly with recipes trained on version 0.1.17 or earlier. (#921)
Fixed a bug where setting fresh = TRUE in prep() wouldn't result in re-prepping the recipe. (#492)
Bug was fixed in step_holiday() which used to error when it was applied to variable with missing values. (#743)
A bug was fixed in step_normalize() which used to error if 1 variable was selected. (#963)
Finally removed step_upsample() and step_downsample() in recipes as they are now available in the themis package.
discretize() and step_discretize() now can return factor levels similar to cut(). (#674)
step_naomit() now actually had their defaults for skip changed to TRUE as was stated in release 0.1.13. (934)
step_dummy() has been made more robust to non-standard column names. (#879)
step_pls() now allows you use use multiple outcomes if they are numeric. (#651)
step_normalize() and step_scale() ignore columns with zero variance, generate a warning and suggest to use step_zv() (#920).
printing for step_impute_knn() now show variables that were imputed instead of variables used for imputing. (#837)
step_discretize() and discretize() will automatically remove missing values if keep_na = TRUE, removing the need to specify keep_na = TRUE and na.rm = TRUE. (#982)
prep() and bake() checks and errors if output of bake.bake_*() isn't a tibble.
step_date() now has a locale argument that can be used to control how the month and dow features are returned. (#1000)
step_nnmf_sparse() uses a different implementation of non-negative matrix factorization that is much faster and enables regularized estimation. (#790)
step_dummy_extract() creates multiple variables from a character variable by extracting elements using regular expressions and counting those elements.
step_filter_missing() can filter columns based on proportion of missingness (#270).
step_percentile() replaces the value of a variable with its percentile from the training set. (#765)
All recipe steps now officially support empty selections to be more aligned with dplyr and other packages that use tidyselect (#603, #531). For example, if a previous step removed all of the columns need for a later step, the recipe does not fail when it is estimated (with the exception of step_mutate()). The documentation in ?selections has been updated with advice for writing selectors when filtering steps are used. (#813)
Fixed bug in step_harmonic() printing and changed defaults to role = "predictor" and keep_original_cols = FALSE (#822).
Improved the efficiency of computations for the Box-Cox transformation (#820).
When a feature extraction step (e.g., step_pca(), step_ica(), etc.) has zero components specified, the tidy() method now lists the selected columns in the terms column.
Deprecation has started for step_nnmf() in favor of step_nnmf_sparse(). (#790)
Steps now have a dedicated subsection detailing what happens when tidy() is applied. (#876)
step_ica() now runs fastICA() using a specific set of random numbers so that initialization is reproducible.
tidy.recipe() now returns a zero row tibble instead of an error when applied to a empty recipe. (#867)
step_zv() now has a group argument. The same filter is applied but looks for zero-variance within 1 or more columns that define groups. (#711)
detect_step() is no longer restricted to steps created in recipes (#869).
New extract_parameter_set_dials() and extract_parameter_dials() methods to extract parameter sets and single parameters from recipe objects.
step_other() now allow for setting threshold = 0 which will result in no othering. (#904)
step_ica() now indirectly uses the fastICA package since that package has increased their R version requirement. Recipe objects from previous versions will error when applied to new data. (#823)
step_kpca*() now directly use the kernlab package. Recipe objects from previous versions will error when applied to new data.
bake() will now error if new_data doesn't contain all the required columns. (#491)
print_step() instead of printer(). This is done for a smoother transition to use cli in the next version. (#871)Added new step_harmonic() (#702).
Added a new step called step_dummy_multi_choice(), which will take multiple nominal variables and produces shared dummy variables. (#716)
The deprecation for step_upsample() and step_downsample() has been escalated from a deprecation warning to a deprecation error; these functions are available in the themis package.
Escalate deprecation for old versions of imputation steps (such as step_bagimpute()) from a soft deprecation to a regular deprecation; these imputation steps have new names like step_impute_bag() (#753).
step_kpca() was un-deprecated and gained the keep_original_cols argument.
The deprecation of the preserve argument to step_pls() and step_dummy() was escalated from a soft deprecation to regular deprecation.
The deprecation of the options argument to step_nzv() was escalated to a deprecation error.
Fix imputation steps for new data that is all NA, and generate a warning for recipes created under previous versions that cannot be imputed with this fix (#719).
A bug was fixed where imputed values via bagged trees would have the wrong levels.
The computations for the Yeo-Johnson transformation were made more efficient (#782).
New recipes_eval_select() which is a developer tool that is useful for creating new recipes steps. It powers the tidyselect semantics that are specific to recipes and supports the modern tidyselect API introduced in tidyselect 1.0.0. Additionally, the older terms_select() has been deprecated in favor of this new helper (#739).
Speed-up/simplification to step_spatialsign()
When only the terms attributes are desired from model.frame use the first row of data to improve speed and memory use (#726).
Use Haversine formula for latitude-longitude pairs in step_geodist() (#725).
Reorganize documentation for all recipe step tidy methods (#701).
Generate warning when user attempts a Box-Cox transformation of non-positive data (@LiamBlake, #713).
step_logit() gained an offset argument for cases where the input is either zero or one (#784)
The tidy() methods for objects from check_new_values(), check_class() and step_nnmf() are now exported.
Added a new step called step_indicate_na(), which will create and append additional binary columns to the data set to indicate which observations are missing (#623).
Added new step_select() (#199).
The threshold argument of step_pca() is now tunable() (#534).
Integer variables used in step_profile() are now kept as integers (and not doubles).
Preserve multiple roles in last_term_info so bake() can correctly respond to has_roles. (#632)
Fixed behavior of the retain flag in prep() (#652).
The tidy() methods for step_nnmf() was rewritten since it was not great (#665), and step_nnmf() now no longer fully loads underlying packages (#685).
Two new selectors that combine role and data type were added: all_numeric_predictors() and all_nominal_predictors(). (#620)
Changed the names of all imputation steps, for example, from step_knnimpute() or step_medianimpute() (old) to step_impute_knn() or step_impute_median() (new) (#614).
Added keep_original_cols argument to several steps:
Added allow_rename argument to recipes_eval_select() (#646).
Performance improvements for step_bs() and step_ns(). The prep() step no longer evaluates the basis functions on the training set and the bake() steps only evaluates the basis functions once for each unique input value (#574)
The neighbors parameter's default range for step_isomap() was changed to be 20-80.
The deprecation for step_upsample() and step_downsample() has been escalated from a soft deprecation to a regular deprecation; these functions are available in the themis package.
Re-licensed package from GPL-2 to MIT. See consent from copyright holders here.
The full tidyselect DSL is now allowed inside recipes step_*() functions. This includes the operators &, |, - and ! and the new where() function. Additionally, the restriction preventing user defined selectors from being used has been lifted (#572).
If steps that drop/add variables are skipped when baking the test set, the resulting column ordering of the baked test set will now be relative to the original recipe specification rather than relative to the baked training set. This is often more intuitive.
More infrastructure work to make parallel processing on Windows less buggy with PSOCK clusters
fully_trained() now returns FALSE when an unprepped recipe is used.
prep() gained an option to print a summary of which columns were added and/or removed during execution.
To reduce confusion between bake() and juice(), the latter is superseded in favor of using bake(object, new_data = NULL). The new_data argument now has no default, so a NULL value must be explicitly used in order to emulate the results of juice(). juice() will remain in the package (and used internally) but most communication and training will use bake(object, new_data = NULL). (#543)
Tim Zhou added a step to use linear models for imputation (#555)
step_filter(), step_slice(), step_sample(), and step_naomit() had their defaults for skip changed to TRUE. In the vast majority of applications, these steps should not be applied to the test or assessment sets.
tidyr version 1.0.0 or later is now required.
step_pls() was changed so that it uses the Bioconductor mixOmics package. Objects created with previous versions of recipes can still use juice() and bake(). With the current version, the categorical outcomes can be used but now multivariate models do not. Also, the new method allows for sparse results.
As suggested by @StefanBRas, step_ica() now defaults to the C engine (#518)
Avoided partial matching on seq() arguments in internal functions.
Improved error messaging, for example when a user tries to prep() a tuneable recipe.
step_upsample() and step_downsample() are soft deprecated in recipes as they are now available in the themis package. They will be removed in the next version.
step_zv() now handles NA values so that variables with zero variance plus are removed.
The selectors all_of() and any_of() can now be used in step selections (#477).
The tune pacakge can now use recipes with check operations (but also requires tune >= 0.1.0.9000).
The tidy method for step_pca() now has an option for returning the variance statistics for each component.
recipes does not directly depend on dials, it has several S3 methods for generics in dials. Version 0.0.5 of dials added stricter validation for these methods, so changes were required for recipes.step_cut() enables you to create a factor from a numeric based on provided break (contributed by Edwin Thoen)yj_trans() to yj_transform() to avoid conflicts.Added flexible naming options for new columns created by step_depth() and step_classdist() (#262).
Small changes for base R's stringsAsFactors change.
Delayed S3 method registration for tune::tunable() methods that live in recipes will now work correctly on R >=4.0.0 (#439, tidymodels/tune#146).
step_relevel() added.
The imputation steps do not change the data type being imputed now. Previously, if the data were integer, the data would be changed to numeric (for some step types). The change is breaking since the underlying data of imputed values are now saved as a list instead of a vector (for some step types).
The data sets were moved to the new modeldata package.
step_num2factor() was rewritten due to a bug that ignored the user-supplied levels (#425). The results of the transform argument are now required to be a function and levels must now be supplied.
Using a minus in the formula to recipes() is no longer allowed (it didn't remove variables anyway). step_rm() or update_role() can be used instead.
When using a selector that returns no columns, juice() and bake() will now return a tibble with as many rows as the original template data or the new_data respectively. This is more consistent with how selectors work in dplyr (#411).
Code was added to explicitly register tunable methods when recipes is loaded. This is required because of changes occurring in R 4.0.
check_class() checks if a variable is of the designated class. Class is either learned from the train set or provided in the check. (contributed by Edwin Thoen)
step_normalize() and step_scale() gained a factor argument with values of 1 or 2 that can scale the standard deviations used to transform the data. (#380)
bake() now produces a tibble with columns in the same order as juice() (#365)
Release driven by changes in tidyr (v 1.0.0).
format_selector()'s wdth argument has been renamed to width
(#250).
step_mutate_at(), step_rename(), and step_rename_at() were added.The use of varying() will be deprecated in favor of an upcoming function tune(). No changes are need in this version, but subsequent versions will work with tune().
format_ch_vec() and format_selector() are now exported (#250).
check_new_values breaks bake if variable contains values that were not observed in the train set (contributed by Edwin Thoen)
When no outcomes are in the recipe, using juice(object, all_outcomes() and bake(object, new_data, all_outcomes() will return a tibble with zero rows and zero columns (instead of failing). (#298). This will also occur when the selectors select no columns.
As alternatives to step_kpca(), two separate steps were added called step_kpca_rbf() and step_kpca_poly(). The use of step_kpca() will print a deprecation message that it will be going away.
step_nzv() and step_poly() had arguments promoted out of their options slot. options can be used in the short term but is deprecated.
step_downsample() will replace the ratio argument with under_ratio and step_upsample() will replace it with over_ratio. ratio still works (for now) but issues a deprecation message.
step_discretize() has arguments moved out of options too; the main arguments are now num_breaks (instead of cuts) and min_unique. Again, deprecation messages are issued with the old argument structure.
Models using the dimRed package (step_kpca(), step_isomap(), and step_nnmf()) would silently fail if the projection method failed. An error is issued now.
Methods were added for a future generic called tunable(). This outlines which parameters in a step can/could be tuned.
Release driven by changes in rlang.
Since 2018, a warning has been issued when the wrong argument was used in bake(recipe, newdata). The depredation period is over and new_data is officially required.
Previously, if step_other() did not collapse any levels, it would still add an "other" level to the factor. This would lump new factor levels into "other" when data were baked (as step_novel() does). This no longer occurs since it was inconsistent with ?step_other, which said that
"If no pooling is done the data are unmodified".
step_normalize() centers and scales the data (if you are, like Max, too lazy to use two separate steps).step_unknown() will convert missing data in categorical columns to "unknown" and update factor levels.If threshold argument of step_other is greater than one then it specifies the minimum sample size before the levels of the factor are collapsed into the "other" category. #289
step_knnimpute() can now pass two options to the underlying knn code, including the number of threads (#323).
Due to changes by CRAN, step_nnmf() only works on versions of R >= 3.6.0 due to dependency issues.
step_dummy() and step_other() are now tolerant to cases where that step's selectors do not capture any columns. In this case, no modifications to the data are made. (#290, #348)
step_dummy() can now retain the original columns that are used to make the dummy variables. (#328)
step_other()'s print method only reports the variables with collapsed levels (as opposed to any column that was tested to see if it needed collapsing). (#338)
step_pca(), step_kpca(), step_ica(), step_nnmf(), step_pls(), and step_isomap() now accept zero components. In this case, the original data are returned.
Small release driven by changes in sample() in the current r-devel.
A new vignette discussing roles has been added.
To provide infrastructure for finalizing varying parameters, an update() method for recipe steps has been added. This allows users to alter information in steps that have not yet been trained.
step_interact will no longer fail if an interaction contains an interaction using column that has been previously filtered from the data. A warning is issued when this happens and no interaction terms will be created.
step_corr was made more fault tolerant for cases where the data contain a zero-variance column or columns with missing values.
Set the embedded environment to NULL in prep.step_dummy to reduce the file size of serialized recipe class objects when using saveRDS.
tidy method for step_dummy now returns the original variable and the levels of the future dummy variables.NA roles of existing columns (#296).Several argument names were changed to be consistent with other tidymodels packages (e.g. dials) and the general tidyverse naming conventions.
K in step_knnimpute was changed to neighbors. step_isomap had the number of neighbors promoted to a main argument called neighbors step_pca, step_pls, step_kpca, step_ica now use num_comp instead of num. , step_isomap uses num_terms instead of num.step_bagimpute moved nbagg out of the options and into a main argument trees.step_bs and step_ns has degrees of freedom promoted to a main argument with name deg_free. Also, step_bs had degree promoted to a main argument.step_BoxCox and step_YeoJohnson had nunique change to num_unique.bake, juice and other functions has newdata changed to new_data. For this version only, using newdata will only result in a wanring.na.rm changed to na_rm.prep and a few steps had stringsAsFactors changed to strings_as_factors.add_role() can now only add new additional roles. To alter existing roles, use update_role(). This change also allows for the possibility of having multiple roles/types for one variable. #221
All steps gain an id field that will be used in the future to reference other steps.
The retain option to prep is now defaulted to TRUE. If verbose = TRUE, the approximate size of the data set is printed. #207
step_integer converts data to ordered integers similar to LabelEncoder #123 and #185step_geodist can be used to calculate the distance between geocodes and a single reference location.step_arrange, step_filter, step_mutate, step_sample, and step_slice implement their dplyr analogs.step_nnmf computes the non-negative matrix factorization for data.rsample function prepper was moved to recipes (issue).step_step_string2factor will now accept factors and leave them as-is.step_knnimpute now excludes missing data in the variable to be imputed from the nearest-neighbor calculation. This would have resulted in some missing data to not be imputed (i.e. return another missing value).step_dummy now produces a warning (instead of failing) when non-factor columns are selected. Only factor columns are used; no conversion is done for character data. issue #186dummy_names gained a separator argument. issue #183step_downsample and step_upsample now have seed arguments for more control over randomness.broom is no longer used to get the tidy generic. These are now contained in the generics package.check_range breaks bake if variable range in new data is outside the range that was learned from the train set (contributed by Edwin Thoen)
step_lag can lag variables in the data set (contributed by Alex Hayes).
step_naomit removes rows with missing data for specific columns (contributed by Alex Hayes).
step_rollimpute can be used to impute data in a sequence or series by estimating their values within a moving window.
step_pls can conduct supervised feature extraction for predictors.
step_log gained an offset argument.
step_log gained a signed argument (contributed by Edwin Thoen).
The internal functions sel2char and printer have been exported to enable other packages to contain steps.
When training new steps after some steps have been previously trained, the retain = TRUE option should be set on previous invocations of prep.
For step_dummy:
one_hot = TRUE option. Thanks to Davis Vaughan.contrast option was removed. The step uses the global option for contrasts.step_other will now convert novel levels of the factor to the "other" level.
step_bin2factor now has an option to choose how the values are translated to the levels (contributed by Michael Levy).
bake and juice can now export basic data frames.
The okc data were updated with two additional columns.
issue 125 that prevented several steps from working with dplyr grouped data frames. (contributed by Jeffrey Arnold)
issue 127 where options to step_discretize were not being passed to discretize.
Edwin Thoen suggested adding validation checks for certain data characteristics. This fed into the existing notion of expanding recipes beyond steps (see the non-step steps project). A new set of operations, called checks, can now be used. These should throw an informative error when the check conditions are not met and return the existing data otherwise.
Steps now have a skip option that will not apply preprocessing when bake is used. See the article on skipping steps for more information.
check_missing will validate that none of the specified variables contain missing data.
detect_step can be used to check if a recipe contains a particular preprocessing operation.
step_num2factor can be used to convert numeric data (especially integers) to factors.
step_novel adds a new factor level to nominal variables that will be used when new data contain a level that did not exist when the recipe was prepared.
step_profile can be used to generate design matrix grids for prediction profile plots of additive models where one variable is varied over a grid and all of the others are fixed at a single value.
step_downsample and step_upsample can be used to change the number of rows in the data based on the frequency distributions of a factor variable in the training set. By default, this operation is only applied to the training set; bake ignores this operation.
step_naomit drops rows when specified columns contain NA, similar to tidyr::drop_na.
step_lag allows for the creation of lagged predictor columns.
step_spatialsign now has the option of removing missing data prior to computing the norm.bake was changed from all_predictors() to everything().verbose option for prep is now defaulted to FALSEstep_dummy was fixed that makes sure that the correct binary variables are generated despite the levels or values of the incoming factor. Also, step_dummy now requires factor inputs.step_dummy also has a new default naming function that works better for factors. However, there is an extra argument (ordinal) now to the functions that can be passed to step_dummy.step_interact now allows for selectors (e.g. all_predictors() or starts_with("prefix") to be used in the interaction formula.step_YeoJohnson gained an na.rm option.dplyr::one_of was added to the list of selectors.step_bs adds B-spline basis functions.step_unorder converts ordered factors to unordered factors.step_count counts the number of instances that a pattern exists in a string.step_string2factor and step_factor2string can be used to move between encodings.step_lowerimpute is for numeric data where the values cannot be measured below a specific value. For these cases, random uniform values are used for the truncated values.step_zv).tidy methods were added for recipes and many (but not all) steps.bake.recipe, the argument newdata is now without a default.bake and juice can now save the final processed data set in sparse format. Note that, as the steps are processed, a non-sparse data frame is used to store the results.First CRAN release.
learn has become prepare and process has become bakestep_lincomb removes variables involved in linear combinations to resolve them.
step_bin2factor)step_regex applies a regular expression to a character or factor vector to create dummy variables.step_dummy and step_interact do a better job of respecting missing values in the data set.recipe objects was changed so that pipes can be used to create the recipe with a formula.process.recipe lost the role argument in factor of a general set of selectors. If no selector is used, all the predictors are returned.