fit() can now take dgCMatrix and sparse tibbles as data
values when add_recipe() or add_variables() is
used (#245, #258).predict() can now take dgCMatrix and sparse tibble
input for new_data argument (#261).extract_fit_time() returns the time it took to train
the workflow (#191).Transition package warnings and errors to use cli instead of rlang (#241).
Minimum R version bumped to 4.0.0.
Added reference to add_variables() in
stages vignette (@brshallo, #190).
While augment.workflow() previously never returned a
.resid column, the method will now return residuals under
the same conditions that augment.model_fit() does
(#201).
augment.workflow() gained an eval_time
argument, enabling augmenting censored regression models (#200,
#213).
The prediction columns are now appended to the LHS rather than
RHS of new_data in augment.workflow(),
following analogous changes in parsnip (#200).
Each of the pull_*() functions soft-deprecated in
workflows v0.2.3 now warn on every usage (#198).
add_recipe() will now error informatively when
supplied a trained recipe (#179).
generics::tune_args() and
generics::tunable() are now registered unconditionally
(#192).Tightens integration with parsnip’s machinery for checking that
needed parsnip extension packages are loaded. add_model()
will now error if a model specification is supplied that requires a
missing extension package (#184).
Introduces support for unsupervised model specifications via the modelenv package (#180).
Simon Couch is now the maintainer (#170).
add_model() now errors if you try to add a model
specification that contains an unknown mode. This is a breaking change,
as previously in some cases it would successfully “guess” the mode. This
change brings workflows more in line with parsnip::fit()
and parsnip::fit_xy() (#160,
tidymodels/parsnip#801).
broom::augment() now works correctly in the edge
case where you had supplied a hardhat blueprint with
composition set to either "matrix" or
"dgCMatrix" (#148).
butcher::axe_fitted() now axes the recipe
preprocessor that is stored inside a workflow, which will reduce the
size of the template data frame that is stored in the
recipe (#147).
add_formula() no longer silently ignores offsets
supplied with offset(). Instead, it now errors at
fit() time with a message that encourages you to use a
model formula through add_model(formula = ) instead
(#162).
New add_case_weights(),
update_case_weights(), and
remove_case_weights() for specifying a column to use as
case weights which will be passed on to the underlying parsnip model
(#118).
R >=3.4.0 is now required, in line with the rest of the tidyverse.
Improved error message in workflow_variables() if
either outcomes or predictors are missing
(#144).
Removed ellipsis dependency in favor of equivalent functions in rlang.
New extract_parameter_set_dials() and
extract_parameter_dials() methods to extract parameter sets
and single parameters from workflow objects.
add_model() and update_model() now use
... to separate the required arguments from the optional
arguments, forcing optional arguments to be named. This change was made
to make it easier for us to extend these functions with new arguments in
the future.
The workflows method for generics::required_pkgs()
is now registered unconditionally (#121).
Internally cleaned up remaining usage of soft-deprecated
pull_*() functions.
workflow() has gained new preprocessor
and spec arguments for adding a preprocessor (such as a
recipe or formula) and a parsnip model specification directly to a
workflow upon creation. In many cases, this can reduce the lines of code
required to construct a complete workflow (#108).
New extract_*() functions have been added that
supersede the existing pull_*() functions. This is part of
a larger move across the tidymodels packages towards a family of generic
extract_*() functions. The pull_*() functions
have been soft-deprecated, and will eventually be removed
(#106).
add_variables() now allows for specifying a bundle
of model terms through add_variables(variables = ),
supplying a pre-created set of variables with the new
workflow_variables() helper. This is useful for supplying a
set of variables programmatically (#92).
New is_trained_workflow() for determining if a
workflow has already been trained through a call to fit()
(#91).
fit() now errors immediately if control
is not created by control_workflow() (#89).
Added broom::augment() and
broom::glance() methods for trained workflow objects
(#76).
Added support for butchering a workflow using
butcher::butcher().
Updated to testthat 3.0.0.
.fit_finalize() for internal usage by the tune
package.New add_variables() for specifying model terms using
tidyselect expressions with no extra preprocessing. For example:
wf <- workflow() %>%
add_variables(y, c(var1, start_with("x_"))) %>%
add_model(spec_lm)One benefit of specifying terms in this way over the formula method
is to avoid preprocessing from model.matrix(), which might
strip the class of your predictor columns (as it does with Date columns)
(#34).
add_formula(),
workflows now uses model-specific information from parsnip to decide
whether to expand factors via dummy encoding (n - 1
levels), one-hot encoding (n levels), or no expansion at
all. This should result in more intuitive behavior when working with
models that don’t require dummy variables. For example, if a parsnip
rand_forest() model is used with a ranger engine, dummy
variables will not be created, because ranger can handle factors
directly (#51, #53).NEWS.md file to track changes to the
package.