Feature Engineering via Dimensionality Reduction — tl_reduce

Use PCA, MDS, or other dimensionality reduction as a preprocessing step for supervised learning. This can improve model performance and interpretability.

Usage

tl_reduce_dimensions(
  data,
  response = NULL,
  method = "pca",
  n_components = NULL,
  ...
)

Arguments

data: A data frame
response: Response variable name (will be preserved)
method: Dimensionality reduction method: "pca" or "mds"
n_components: Number of components to retain, at most the number the method computes: one per numeric column for PCA, and k (default 2, passed through ...) for MDS. NULL keeps them all.
...: Additional arguments for the dimensionality reduction method

Value

A list with components:

data: The transformed data frame with reduced-dimension columns and the response variable (if provided).
reduction_model: The fitted tidylearn dimensionality reduction model.
original_data: The original input data frame.
response: The response variable name, or NULL.

Examples

# \donttest{
# Reduce dimensions before classification
reduced <- tl_reduce_dimensions(
  iris, response = "Species",
  method = "pca", n_components = 3
)
model <- tl_model(reduced$data, Species ~ ., method = "tree")
# }