Hybrid Penalized Partial Least Squares for Mixed Data

Fits Penalized Partial Least Squares (PLS) regression when predictors are hybrid objects that combine functional curves (infinite-dimensional 'fda' objects) and scalar covariates (finite-dimensional numeric matrices). The package treats a hybrid predictor as an element of a product Hilbert space formed by the functional and Euclidean components, and implements the arithmetic (addition, scalar multiplication, and inner products, including roughness-penalized inner products) needed to run penalized PLS directly in that space. The algorithm extracts latent components that maximize covariance with a scalar response while penalizing roughness of the estimated functional coefficient curves. Helpers are included for constructing hybrid predictors, two-step within- and between-modality normalization, train/test splitting, synthetic data generation, cross-validated component selection, and prediction. The method is described in Mun and Jang (2026) .

FSHybridPLS

Hybrid Penalized Partial Least Squares for predictors that combine functional curves (fda::fd) and scalar covariates.

The method is described in Mun and Jang (2026):
https://doi.org/10.48550/arXiv.2601.16364

Installation

# Once on CRAN:
# install.packages("FSHybridPLS")

# Development version:
# remotes::install_github("Jong-Min-Moon/FShybridPLS")

Quick start

library(FSHybridPLS)
set.seed(1)

sim <- simulate_hybrid_data(n = 60, n_functional = 1, n_scalar = 3, n_basis = 7)
prep <- split_and_normalize_all(sim$W, sim$y, train_ratio = 0.7)

fit <- fit_hybridPLS(
  prep$predictor_train,
  prep$response_train,
  n_iter = 3,
  lambda = 1e-3,
  validation_data = list(
    W_test = prep$predictor_test,
    y_test = prep$response_test
  )
)

fit
preds <- predict(fit, prep$predictor_test, n_components = fit$n_iter)
sqrt(mean((prep$response_test - preds)^2))

Main API

Citation

citation("FSHybridPLS")