bolasso: Bootstrapped LASSO (Bolasso) for regression and...
In dmolitor/umbrella: A conglomeration of modeling and data munging functionality

Description Usage Arguments Details Value Examples

bolasso implements Bootstrapped LASSO regression and classification.

bolasso(
  train_df,
  nboot,
  formula,
  family,
  test_df = NULL,
  predict_df = NULL,
  free_vars = NULL,
  nfold = NULL,
  lambda = "min",
  sparsity_threshold = NULL,
  selection_threshold = 0.9,
  verbose = FALSE,
  ...
)

`train_df`	An input dataframe with `y` and `X`.
`nboot`	A numeric value specifying the number of bootstrap replicates.
`formula`	A formula for the covariates.
`family`	A string that specifies either 'gaussian' or 'binomial'.
`test_df`	A dataframe containing the same columns as `train_df`. The training set.
`predict_df`	A dataframe matching `train_df`. This is to generate predictions using the trained & tested model. This argument is optional.
`free_vars`	A string or character vector specifying which covariate(s) to never penalize. This argument is optional.
`nfold`	The number of cross-validation folds. Only specify if cross-validation is desired. This argument is optional.
`lambda`	A string specifying which lambda to use for prediction when utilizing cross-validation. Typically either "min" or "1se". Default value is "min".
`sparsity_threshold`	A numeric value in [0, 1]. Any variable with a percentage of sparsity greater than this value will be dropped. This argument is optional.
`selection_threshold`	A numeric value in [0, 1]. Variables must be selected in at least this percentage of bootstrap replicates. Default value is 0.9.
`verbose`	Logical indicating whether to return progress statements. Default is `FALSE`.
`...`	Generic argument to which you can pass any other valid gamlr argument, such as `standardize = FALSE`.

The bolasso function bootstraps LASSO regression, as found in the gamlr package, across n bootstrap partitions. This achieves more robust variable selection than standard LASSO and improves predictive accuracy. It handles standard OLS regression and binomial logistic regression.

A list containing the LASSO model, predicted_values, residuals, and selected variables.

## Not run: 
idx <- train_test_validate(iris$Sepal.Length, train.p = .6, test.p = .2)

initialize_parallel()

bolasso_model <- bolasso(train_df = iris[idx$train, ],
                         nboot =100,
                         formula = Sepal.Length ~ .,
                         family = "gaussian",
                         test_df = iris[idx$test, ],
                         predict_df = iris[idx$validate, ],
                         nfold = 5,
                         verbose = TRUE)

## End(Not run)