analyzeImportanceFeaturesFBM: Analyze and Plot Feature Importance for Feature-Based Models...

In predomics/predomicspkg: Interpretable Prediction in Omics Data

Analyze and Plot Feature Importance for Feature-Based Models (FBM)

Description

This function analyzes and visualizes feature importance for FBM models, creating plots for feature prevalence, importance, and effect size. It supports both single and multiple experiments, handling classification and regression modes.

Usage

analyzeImportanceFeaturesFBM(
  clf_res,
  X,
  y,
  makeplot = TRUE,
  saveplotobj = TRUE,
  name = "",
  verbose = TRUE,
  pdf.dims = c(width = 25, height = 20),
  filter.cv.prev = 0.25,
  nb.top.features = 100,
  scaled.importance = FALSE,
  k_penalty = 0.75/100,
  k_max = 0
)

Arguments

clf_res	A classifier result object or a list of classifier results. Each classifier result must contain the trained models and the classifier parameters.
X	The feature matrix.
y	The response variable.
makeplot	Logical, if 'TRUE' (default), plots will be generated and displayed.
saveplotobj	Logical, if 'TRUE' (default), the plot objects will be saved as an RData file.
name	A character string for the output file name prefix.
verbose	Logical, if 'TRUE' (default), progress messages are printed.
pdf.dims	A numeric vector specifying the width and height of the PDF output.
filter.cv.prev	Numeric, a cutoff for the cross-validation prevalence filter.
nb.top.features	Numeric, the number of top features to display.
scaled.importance	Logical, if 'TRUE', scales feature importance values.
k_penalty	Numeric, a penalty factor applied to control model selection based on sparsity.
k_max	Numeric, maximum number of features to include.

Details

This function examines the FBM classifier results, retrieves the best population of models, calculates feature importance, and generates plots to visualize feature prevalence, importance, and effect size. If multiple experiments are provided, results are averaged across experiments.

**Generated Plots**: - **Feature Prevalence (FBM)**: Shows the frequency and orientation of features across models. - **Feature Importance**: Visualizes feature importance scores. - **Effect Sizes**: Displays the effect sizes of features.

For classification, Cliff's delta is used for effect sizes, and Spearman's rho is used for regression.

Value

If 'makeplot' is 'TRUE', a combined plot of feature importance, prevalence, and effect sizes is returned. Otherwise, a list of individual plot objects.

Examples

## Not run: 
analyzeImportanceFeaturesFBM(clf_res = my_clf_result, X = X_data, y = y_data, name = "example_analysis")

## End(Not run)

predomics/predomicspkg documentation built on Dec. 11, 2024, 11:06 a.m.