[0.2.0]
- Non-Development version of 0.2.0.
- Minor internal code changes.
- Additional testthat tests to assess outputs.
- Removes explicit roxygen2 export of internal private functions.
- Documents the
contr.dummy
function from thekknn
package sincetrain.kknn
requires this function to be in the
current namespace to work. - Allows additional arguments for multiple algorithms to be used.
[0.2.0.9000] [Development]
♻ Changed
- Refactored package internally to make code more reusable and maintainable. Version is in still in development but has
passed previous testthat tests and all functions can be used. - Some parameters for
classCV
andgenFolds
have been grouped together. For instance,split
,n_folds
,standardize
,
remove_obs
,random_seed
,stratified
, etc are no longer separate input parameters. They are now apart of the new
train_params
parameter as elements (e.gtrain_params = list(split = 0.8, n_folds = 5, standardize = TRUE))
. Additionally,
model_params
,save
, andparallel_configs
were also created to group similar parameters in the formerclassCV
input parameters. For all functions,model_type
is notmodels
and forprint
,parameters
is nowconfigs
. classCV
output object is more organized and includes "configs" for user specified arguments and model-specific arguments,
"class_summary" for information pertaining to classes such as the names of the classes, indices, proportions,
"data_partitions" to include the indices, class proportions in each split/fold, and dataframes if requested, "imputation"
for imputation information, "models" if models are requested, and "metrics" for metrics.- Can request final model with having to specify
n_folds
orsplit
.
🐛 Fixes
-
The previous behavior were observations that are missing the target or excluded; however in addition to this,
when imputation is requested, the target variable is excluded from being a predictor for imputation. -
Prior to imputation, regardless if standardization is requested, all numerical columns are standardized.
-
Error when saving plots in RStudio.
-
Metrics for latest GBM version should no longer produce NAs
-
[RE-UPLOAD]: Version that allows final model to run without specifying
split
orn_folds
,
also includes additional tests. Still has the same version number - 0.2.0.9000. Re-upload also includes a fix for
logistic regression, sincemodel_params$threshold
was called instead ofmodel_params$logistic_threshold
, resulting in
the logistic threshold to be NULL and the prediction vector empty.