bolasso 0.5.0

• Addresses [#19]. When family = "multinomial" is
  specified, previously bolasso did not correctly support variable selection, plotting, etc.
  This version addresses this. In general, when family = "multinomial", all outputs become a list
  where each list element corresponds to a unique outcome in the multinomial response vector. However,
  the content of each list element will be the same as it originally was (e.g. a tibble of
  coefficients, or a vector of variable names, etc.). In the case of plotting, it will create a
  facetted plot with one facet per unique outcome. The only limitations are that family = "mgaussian"
  is not currently supported, and when predicting with family = "multinomial", only type = "class"
  is currently supported. These may be added at a later date.

bolasso 0.4.0

• Allows the user to extract the bootstrap indices with bootstrap_samples().

  library(bolasso)
  
  model <- bolasso(mpg ~ hp + wt, data = mtcars, n.boot = 10)
  bootstrap_samples(model)
  

• Fixes #12

• Fixes #13

bolasso 0.3.0

New Features

• Fast Estimation Mode:

  • bolasso() gains a fast argument which optimizes computation by using a single cross-validated regression on the entire dataset to determine the optimal regularization parameter (lambda). This approach bypasses the need for cross-validation within each bootstrap replicate, drastically reducing computation time, especially beneficial for large datasets or when using a high number of bootstrap replicates.

    # Fast mode reduces computation time by using a single cross-validated lambda
    model_fast <- bolasso(
    diabetes ~ .,
    data = train,
    n.boot = 1000,
    progress = FALSE,
    family = "binomial",
    fast = TRUE
    )

• Enhanced Variable Selection Methods:

  • selected_vars() is now a shorthand for selected_variables().
  • selected_variables()/selected_vars() supports two variable selection algorithms via the method argument: the Variable Inclusion Probability (VIP) method and the Quantile (QNT) method. The VIP method selects variables that appear in a high percentage of bootstrap models, while the QNT method selects variables based on bootstrap confidence intervals. Set method = "vip" or method = "qnt", respectively.

    # Select variables using the VIP method with a 95% threshold
    selected_vars_vip <- selected_variables(model, threshold = 0.95, method = "vip")
    
    # Select variables using the QNT method
    selected_vars_qnt <- selected_variables(model, threshold = 0.95, method = "qnt")

• Tidy Method for Bolasso Objects:

  • tidy() extracts a tidy tibble summarizing bootstrap-level coefficients for each covariate. This method provides a clean and organized way to inspect model coefficients.

    # Extract a tidy tibble of coefficients
    tidy_coefs <- tidy(model, select = "lambda.min")

• Variable Selection Visualization:

  • plot_selection_thresholds() provides a visual representation of the selection thresholds for each variable. This visualization helps users understand the stability and robustness of variable selection across different thresholds and methods.

    # Visualize selection thresholds for variables
    plot_selection_thresholds(model, select = "lambda.min")

Improvements

• Plotting Coefficient Distributions:

  • plot_selected_variables() visualizes the coefficient distributions for only the selected variables. This function provides a focused view of the most relevant variables in the model.

    # Plot coefficient distributions for selected variables
    plot_selected_variables(
    model,
    threshold = 0.95,
    method = "vip",
    select = "lambda.min"
    )

  • plot visualizes the coefficient distributions for all model covariates.

    # Plot coefficient distributions for selected variables
    plot(model, select = "lambda.min")

Bolasso v0.2.0

This version implements several minor enhancements and bug fixes. See NEWS.md for full description of changes.

Bolasso v0.1.0

Initial Bolasso CRAN release

• v.0.1.0