- Add additional arguments to
synthesize()for saving incomplete syntheses. - Redesign
postsynthto inherit frompresynthfor retaining incomplete synthesis specifications. - Add
postsynth_to_roadmap()andpostsynth_to_synth_spec()for restarting incomplete syntheses.
- Prepare tidysynthesis for CRAN release (#26).
- Add rounding to
noise()(#6) and fixobs_per_ntileinnoise()(#3). - Add default noise mechanism in
synth_spec()for when exactly one type of noise mechanism is specified (#2). - Update printing for
synthesize()(#4) and constraints (#10). - Add a gini method to
add_sequence_factor()and resolve minor bugs (#15, #16). - Update documentation for built-in data sets (#7).
- Add
constraints()support for generic conditions and categorical constraints. - Update
noise()to accept generic functions and parameters. - Add new
add_noise_*()functions for use with new genericnoise(). - Audit and update all user-facing error messages for more informative errors.
- Refactor
synthesize()internals for more modular development and validation. - New
replicates()logic for replicating start methods and conditional syntheses independently.
- New
tidymodelsinspired API. - Add
visit_sequenceAPI for mixed-type data-dependent sequencing. - Add
start_methodfor randomizedstart_datatransformations. - Add
tunersand user-specified hyperparameter tuning. - Update
constraints()to work withNAvalues. - Remove dependence on
furrrand stratified synthesis.
- Fix a bug caused by
col_schemainvisit_sequencenot following the sort order of thevisit_sequence.
- Add ability to synthesize numeric and categorical variables with
NA. - Add ability to extract information from synthesizer models using
extract_*()functions. - Refactor
schema()object and reverse the order ofschema()andvisit_sequence()in a typical synthesis. - Improve
ldiversitycalculations for tree-based models. - Fix issue with empty factor levels caused by new version of
library(recipes).
- Refactor
construct_recipes()to handle much larger data sets. The.notation in the formula caused a C Stack error for with large visit sequences. - Add variable-by-variable hyperparameter tuning.
- Add ability to store model metrics from each predictive model.
- Add new schema object to support mixed-data types and data validation.
- Add categorical sampler for random forests and add samplers for logistic regression and Poisson regression.
- Improve validators for S3 objects.
- Add
library(ranger). - Add support for batch synthesis.
- Update installation instruction.
- Fix bug in
sample_rpart()for categorical synthesis. - Fix bug in progress bar caused by
library(furrr). The package now useslibrary(progressr)for tracking progress.
- Add default objects in
presynth()fornoise,constraints, andreplicatessonoise(),constraints(), andreplicates()don't always need to be called. - Add code to handle empty factor levels, which break many models.
- Add tests in
roadmap()for ordinal variables. - Make the
"identity"method use the same structure as models fromlibrary(parsnip). - Add tests and fix all issues with s3 print methods.
- Rewrite joins with
multiple = "all"to minimize R warnings.
- Create first numbered version of
library(tidysynthesis)! - Rewrite
assign_constraints()andsynthesize_j()to improve performance by as much as 10x. - Factors are now synthesized as factors instead of strings.
- Add "identity" methods for variables without variation. This previously caused decision trees to error out.
- Add more tests for the
ldiversitytibble. - Add defaults for
noise,constraints, andreplicatesinpresynth().