Title: SelectBoost-Style Variable Selection for Functional Data Analysis
Date: 2026-04-02
Version: 0.5.0
Author: Frederic Bertrand ORCID iD [cre, aut]
Maintainer: Frederic Bertrand <frederic.bertrand@lecnam.net>
Description: Implements 'SelectBoost'-style variable selection workflows for functional data analysis. The package provides FDA-native design and preprocessing objects for raw curves, spline-basis expansions, Functional principal component analysis scores, and scalar covariates; grouped stability-selection routines based on repeated subject-level subsampling; multiple selector backends including lasso, group lasso, and sparse-group lasso; FDA-aware grouping functions and calibration helpers for 'SelectBoost'; method-comparison utilities; a formula interface; simulation, benchmarking, and validation helpers with mapped ground truth; targeted sensitivity-study utilities and shipped benchmark summaries for mean 'F1' comparisons between FDA-aware and plain 'SelectBoost' workflows; small example datasets; and an optional adapter to the native stability-selection interface from the 'FDboost' package.
LazyData: true
License: GPL-3
Encoding: UTF-8
RoxygenNote: 7.3.3
Depends: R (≥ 2.10)
Imports: SelectBoost, stats
Suggests: FDboost, glmnet, grpreg, knitr, pkgdown, rmarkdown, SGL, stabs, testthat (≥ 3.0.0)
VignetteBuilder: knitr
Config/testthat/edition: 3
Config/Needs/website: pkgdown
URL: https://fbertran.github.io/SelectBoost.FDA/, https://github.com/fbertran/SelectBoost.FDA
BugReports: https://github.com/fbertran/SelectBoost.FDA/issues
NeedsCompilation: no
Packaged: 2026-04-02 23:01:50 UTC; bertran7
Repository: CRAN
Date/Publication: 2026-04-10 13:30:09 UTC

Apply an FDA Preprocessor

Description

Applies a fitted preprocessor to new functional predictors and optional scalar covariates, returning an fda_matrix object compatible with the selection routines.

Usage

apply_fda_preprocessor(object, predictors, scalar_covariates = NULL, ...)

Arguments

object

A fitted fda_preprocessor.

predictors

New functional predictors.

scalar_covariates

Optional scalar covariates.

...

Not used.

Value

An object of class fda_matrix.

Flatten Functional Predictors Into a Matrix

Description

Accepts a standard numeric matrix/data frame or a named list of functional blocks. List inputs are column-bound while preserving the original block membership of each coefficient, which is later reused for grouped stability selection and FDA-aware SelectBoost grouping.

Usage

as_functional_matrix(x, center = FALSE, scale = FALSE)

Arguments

x

A numeric matrix/data frame, an fda_grid, an fda_basis, an fda_design, or a list of such objects. Each list element is treated as one functional block.

center, scale

Passed to base::scale() when either argument is TRUE.

Value

An object of class fda_matrix with elements x, blocks, and positions.

Benchmark FDA Selection Methods on Shared Ground Truth

Description

Runs compare_selection_methods() on a simulated dataset and evaluates the fitted objects against the mapped truth.

Usage

benchmark_selection_methods(
  data,
  methods = c("stability", "interval", "selectboost", "plain_selectboost"),
  levels = c("feature", "group"),
  stability_args = list(),
  interval_args = list(),
  selectboost_args = list(),
  plain_selectboost_args = list(),
  fdboost_model = NULL,
  fdboost_args = list(),
  keep_comparison = TRUE
)

Arguments

data

An object returned by simulate_fda_scenario().

methods

Methods passed to compare_selection_methods().

levels

Evaluation levels.

stability_args, interval_args, selectboost_args, plain_selectboost_args

Additional arguments passed to compare_selection_methods().

fdboost_model, fdboost_args

Optional FDboost inputs forwarded to compare_selection_methods().

keep_comparison

Should the fitted comparison object be stored?

Value

An object of class fda_benchmark.

Examples

sim <- simulate_fda_scenario(n = 24, grid_length = 16, seed = 1)
bench <- benchmark_selection_methods(
  sim,
  methods = c("selectboost", "plain_selectboost"),
  selectboost_args = list(B = 3, steps.seq = 0.5, c0lim = FALSE),
  plain_selectboost_args = list(B = 3, steps.seq = 0.5, c0lim = FALSE)
)
head(bench$metrics)

Calibrate Interval Widths

Description

Runs interval stability selection over candidate interval widths.

Usage

calibrate_interval_width(
  design,
  widths,
  step = NULL,
  overlap = FALSE,
  selector = "lasso",
  keep_fits = FALSE,
  seed = NULL,
  ...
)

Arguments

design

An fda_design object.

widths

Candidate interval widths.

step

Optional step size. Defaults to widths.

overlap

Should the interval groups overlap?

selector

Base selector passed to interval_stability_selection().

keep_fits

Should the fitted objects be stored in the result?

seed

Optional seed used to create deterministic per-grid seeds.

...

Additional arguments passed to interval_stability_selection().

Value

An object of class fda_calibration_grid.

Calibrate SelectBoost c0 Values

Description

Runs FDA-SelectBoost on a user-provided or suggested c0 grid.

Usage

calibrate_selectboost(
  design,
  selector = "msgps",
  c0_grid = NULL,
  grid_method = c("quantile", "linear"),
  association_method = c("correlation", "neighborhood", "hybrid", "interval"),
  keep_fit = TRUE,
  ...
)

Arguments

design

An fda_design object.

selector

Base selector passed to fit_selectboost().

c0_grid

Optional explicit c0 grid.

grid_method

Rule used by suggest_c0_grid() when c0_grid is omitted.

association_method

Passed to suggest_c0_grid() and fit_selectboost().

keep_fit

Should the fitted object be stored in the result?

...

Additional arguments passed to fit_selectboost().

Value

An object of class fda_calibration_grid.

Calibrate Stability-Selection Parameters

Description

Runs grouped stability selection over a grid of subsampling fractions and cutoff values.

Usage

calibrate_stability_selection(
  design,
  selector = "group_lasso",
  sample_fraction_grid = c(0.5, 0.632, 0.75),
  cutoff_grid = c(0.6, 0.75, 0.9),
  keep_fits = FALSE,
  seed = NULL,
  ...
)

Arguments

design

An fda_design object.

selector

Base selector passed to fit_stability().

sample_fraction_grid

Candidate subsampling fractions.

cutoff_grid

Candidate cutoff values.

keep_fits

Should the fitted objects be stored in the result?

seed

Optional seed used to create deterministic per-grid seeds.

...

Additional arguments passed to fit_stability().

Value

An object of class fda_calibration_grid.

Compare FDA Selection Methods

Description

Runs multiple selection workflows on the same fda_design object and returns both the fitted objects and a comparison table.

Usage

compare_selection_methods(
  design,
  methods = c("stability", "interval", "selectboost"),
  stability_args = list(),
  interval_args = list(),
  selectboost_args = list(),
  plain_selectboost_args = list(),
  fdboost_model = NULL,
  fdboost_args = list()
)

Arguments

design

An fda_design object.

methods

Methods to run. Supported values are "stability", "interval", "selectboost", "plain_selectboost", and "fdboost".

stability_args, interval_args, selectboost_args, plain_selectboost_args

Named lists of arguments passed to the corresponding fitting functions.

fdboost_model

Optional fitted FDboost object used when methods includes "fdboost".

fdboost_args

Additional arguments passed to fdboost_stability_selection().

Value

An object of class fda_method_comparison.

Examples

sim <- simulate_fda_scenario(n = 24, grid_length = 16, seed = 1)
comparison <- compare_selection_methods(
  sim$design,
  methods = c("selectboost", "plain_selectboost"),
  selectboost_args = list(B = 3, steps.seq = 0.5, c0lim = FALSE),
  plain_selectboost_args = list(B = 3, steps.seq = 0.5, c0lim = FALSE)
)
summary(comparison)

Evaluate Selection Recovery Against Ground Truth

Description

Computes support-recovery metrics for fitted selection objects against the truth generated by simulate_fda_scenario().

Usage

evaluate_selection(x, truth, level = c("feature", "group", "basis"), ...)

Arguments

x

A fitted selection object or an fda_method_comparison.

truth

Ground-truth object, typically the value returned by simulate_fda_scenario().

level

Evaluation level: "feature", "group", or "basis".

...

Additional arguments passed to the relevant method.

Value

A data frame with recovery metrics.

Basis-Expanded Functional Predictor

Description

Constructor for a functional predictor represented by basis coefficients or FPCA scores.

Usage

fda_basis(
  coefficients,
  basis_type = c("generic", "spline", "wavelet", "fpca"),
  argvals = NULL,
  component_names = NULL,
  name = NULL,
  unit = NULL
)

Arguments

coefficients

Numeric matrix with one row per observation.

basis_type

Label describing the representation.

argvals

Optional labels or positions for basis functions/components.

component_names

Optional names for coefficient columns.

name

Optional predictor name.

unit

Optional unit for the basis domain.

Value

An object of class fda_basis.

Spline-Basis Preprocessing Spec

Description

Spline-Basis Preprocessing Spec

Usage

fda_bspline(
  df = 6L,
  degree = 3L,
  intercept = TRUE,
  center = FALSE,
  scale = FALSE
)

Arguments

df

Degrees of freedom used by splines::bs().

degree

Spline degree.

intercept

Should the spline basis include an intercept column?

center, scale

Logical flags controlling column-wise centering and scaling of the resulting coefficients.

Value

An object of class fda_preprocess_spec.

Functional Design Object

Description

Bundles the response, functional predictors, family, and a reversible feature map. This is the FDA-native entry point for the higher-level fitting functions.

Usage

fda_design(
  response = NULL,
  predictors,
  scalar_covariates = NULL,
  family = c("gaussian", "binomial"),
  id = NULL,
  center = FALSE,
  scale = FALSE,
  transforms = NULL,
  scalar_transform = NULL,
  preprocessor = NULL
)

Arguments

response

Response vector.

predictors

A single predictor or a named list of predictors. Elements can be fda_grid, fda_basis, fda_scalar, matrices, data frames, or numeric vectors.

scalar_covariates

Optional scalar covariates supplied separately from the functional predictors.

family

Model family.

id

Optional observation identifiers.

center, scale

Backward-compatible shortcuts for applying an identity transform with centering and scaling to the functional predictors.

transforms

Optional preprocessing specs for the functional predictors.

scalar_transform

Optional preprocessing specs for scalar covariates.

preprocessor

Optional fitted fda_preprocessor. When supplied, it is reused instead of fitting preprocessing from the current data.

Value

An object of class fda_design.

Examples

data("spectra_example", package = "SelectBoost.FDA")
idx <- 1:20
design <- fda_design(
  response = spectra_example$response[idx],
  predictors = list(
    signal = fda_grid(
      spectra_example$predictors$signal[idx, ],
      argvals = spectra_example$grid,
      name = "signal"
    ),
    nuisance = fda_grid(
      spectra_example$predictors$nuisance[idx, ],
      argvals = spectra_example$grid,
      name = "nuisance"
    )
  ),
  scalar_covariates = spectra_example$scalar_covariates[idx, ],
  scalar_transform = fda_standardize(),
  family = "gaussian"
)
summary(design)

Build an FDA Design from a Formula

Description

Supports additive formulas of the form y ~ signal + noise + age + batch, where functional terms are supplied as matrices, fda_grid, or fda_basis objects in data, and scalar terms are expanded through stats::model.matrix().

Usage

fda_design_formula(
  formula,
  data,
  family = c("gaussian", "binomial"),
  transforms = NULL,
  scalar_transform = NULL,
  preprocessor = NULL,
  center = FALSE,
  scale = FALSE
)

Arguments

formula

An additive formula with a single response.

data

A list or data frame containing the variables referenced in formula.

family, transforms, scalar_transform, preprocessor, center, scale

Passed to fda_design().

Value

An object of class fda_design.

FPCA Preprocessing Spec

Description

FPCA Preprocessing Spec

Usage

fda_fpca(
  n_components = 3L,
  variance_explained = NULL,
  center = TRUE,
  scale = FALSE
)

Arguments

n_components

Number of principal components to retain.

variance_explained

Optional cumulative explained variance target in ⁠(0, 1]⁠. When supplied, it overrides n_components.

center, scale

Passed to stats::prcomp().

Value

An object of class fda_preprocess_spec.

Functional Predictor on a Common Grid

Description

Constructor for one discretized functional predictor sampled on a common grid.

Usage

fda_grid(values, argvals = NULL, name = NULL, unit = NULL)

Arguments

values

Numeric matrix with one row per observation.

argvals

Optional vector of grid values. Defaults to seq_len(ncol(values)).

name

Optional predictor name.

unit

Optional unit for the grid axis.

Value

An object of class fda_grid.

Identity Preprocessing Spec

Description

Identity Preprocessing Spec

Usage

fda_identity(center = FALSE, scale = FALSE)

Arguments

center, scale

Logical flags controlling column-wise centering and scaling of the transformed features.

Value

An object of class fda_preprocess_spec.

Scalar Predictor Constructor

Description

Wraps scalar covariates so they can participate in the same feature-mapping and preprocessing machinery as functional predictors.

Usage

fda_scalar(values, name = NULL, unit = NULL)

Arguments

values

Numeric vector or matrix with one row per observation.

name

Optional predictor name.

unit

Optional unit label.

Value

An object of class fda_scalar.

Standardization Preprocessing Spec

Description

Standardization Preprocessing Spec

Usage

fda_standardize(center = TRUE, scale = TRUE)

Arguments

center, scale

Logical flags controlling column-wise centering and scaling. Both default to TRUE.

Value

An object of class fda_preprocess_spec.

Stability Selection for FDboost Fits

Description

Thin adapter to the stabsel.FDboost() method. This is the native route when the model itself is already fitted with FDboost.

Usage

fdboost_stability_selection(model, ...)

Arguments

model

A fitted FDboost object.

...

Additional arguments forwarded to stabs::stabsel().

Value

A stabsel object.

Fit an FDA Preprocessor

Description

Learns train/test-safe preprocessing transforms for functional predictors and optional scalar covariates. The fitted object can be reused to create compatible fda_design objects on new data.

Usage

fit_fda_preprocessor(
  predictors,
  scalar_covariates = NULL,
  transforms = NULL,
  scalar_transform = NULL
)

Arguments

predictors

One predictor or a named list of predictors.

scalar_covariates

Optional scalar covariates supplied as a vector, matrix/data frame, fda_scalar, or a named list.

transforms

Optional preprocessing specs for functional predictors.

scalar_transform

Optional preprocessing specs for scalar covariates.

Value

An object of class fda_preprocessor.

Fit SelectBoost from an FDA Design

Description

Fit SelectBoost from an FDA Design

Usage

fit_selectboost(design, ...)

Arguments

design

An fda_design object.

...

Additional arguments forwarded to selectboost_fda().

Value

An object inheriting from selectboost_fda_result.

Examples

sim <- simulate_fda_scenario(n = 24, grid_length = 16, seed = 1)
fit <- fit_selectboost(
  sim$design,
  mode = "fast",
  steps.seq = 0.5,
  c0lim = FALSE,
  B = 3
)
head(selection_map(fit, c0 = colnames(fit$feature_selection)[1]))

Fit FDA-SelectBoost from a Formula

Description

Fit FDA-SelectBoost from a Formula

Usage

fit_selectboost_formula(
  formula,
  data,
  family = c("gaussian", "binomial"),
  transforms = NULL,
  scalar_transform = NULL,
  preprocessor = NULL,
  center = FALSE,
  scale = FALSE,
  ...
)

Arguments

formula, data, family, transforms, scalar_transform, preprocessor, center, scale

Passed to fda_design_formula().

...

Additional arguments passed to fit_selectboost().

Value

An object inheriting from selectboost_fda_result.

Fit Grouped Stability Selection from an FDA Design

Description

Fit Grouped Stability Selection from an FDA Design

Usage

fit_stability(design, ...)

Arguments

design

An fda_design object.

...

Additional arguments forwarded to stability_selection_fda().

Value

An object inheriting from fda_stability_selection.

Examples

sim <- simulate_fda_scenario(n = 24, grid_length = 16, seed = 1)
if (requireNamespace("glmnet", quietly = TRUE)) {
  fit <- fit_stability(
    sim$design,
    selector = "lasso",
    B = 4,
    cutoff = 0.4,
    seed = 1
  )
  head(selection_map(fit))
}

Fit Stability Selection from a Formula

Description

Fit Stability Selection from a Formula

Usage

fit_stability_formula(
  formula,
  data,
  family = c("gaussian", "binomial"),
  transforms = NULL,
  scalar_transform = NULL,
  preprocessor = NULL,
  center = FALSE,
  scale = FALSE,
  ...
)

Arguments

formula, data, family, transforms, scalar_transform, preprocessor, center, scale

Passed to fda_design_formula().

...

Additional arguments passed to fit_stability().

Value

An object inheriting from fda_stability_selection.

Functional Association Matrix

Description

Computes or post-processes an absolute association matrix for discretized or basis-expanded functional predictors.

Usage

functional_association(
  x,
  association = NULL,
  method = c("correlation", "neighborhood", "hybrid", "interval"),
  within_blocks = TRUE,
  bandwidth = NULL,
  interval_groups = NULL,
  width = NULL,
  step = width,
  decay = 1
)

Arguments

x

Any input accepted by as_functional_matrix().

association

Optional square association matrix supplied by the user. When omitted, abs(stats::cor(X)) is used.

method

Association structure. "correlation" uses the absolute correlation matrix, "neighborhood" uses local positional similarity, "hybrid" multiplies correlation by a neighborhood kernel, and "interval" induces associations within interval groups.

within_blocks

Should cross-block associations be zeroed out?

bandwidth

Optional maximum within-block lag retained in the association matrix.

interval_groups

Optional interval grouping used when method = "interval".

width, step

Interval parameters used when method = "interval" and interval_groups is omitted.

decay

Positive exponent controlling the neighborhood kernel.

Value

A square absolute association matrix with unit diagonal.

Block-Level Groups for Functional Predictors

Description

Returns one group label per column, with each functional block defining a group.

Usage

functional_block_groups(x)

Arguments

x

Any input accepted by as_functional_matrix().

Value

An integer vector of group memberships.

Interval Groups for Discretized Functional Predictors

Description

Creates non-overlapping interval groups within each functional block. This is useful when one wants region-level stability summaries instead of pointwise selection frequencies.

Usage

functional_interval_groups(x, width, step = width, overlap = FALSE)

Arguments

x

Any input accepted by as_functional_matrix().

width

Positive integer interval width within each block.

step

Step size between interval starts. Only non-overlapping intervals are supported by default.

overlap

Logical; should intervals be allowed to overlap? When TRUE, the result is returned as an overlapping group structure.

Value

Either an integer group vector with an interval_table attribute or an overlapping group structure of class fda_group_list.

Interval Stability Selection

Description

Convenience wrapper around stability_selection_fda() that first creates non-overlapping interval groups within each functional block.

Usage

interval_stability_selection(
  x,
  y = NULL,
  width,
  step = width,
  overlap = FALSE,
  ...
)

Arguments

x

Any input accepted by as_functional_matrix(), or an fda_design object.

y

Response vector. Leave as NULL when x is an fda_design.

width

Positive interval width.

step

Step size between interval starts.

overlap

Logical; should the interval groups overlap?

...

Additional arguments forwarded to stability_selection_fda().

Value

An object of class fda_interval_stability_selection.

FDA-Aware Grouping Function for SelectBoost

Description

Builds a closure that can be passed directly to ⁠group=⁠ in SelectBoost::fastboost() or SelectBoost::autoboost(). The returned grouping function respects functional block boundaries and can optionally restrict groups to local neighborhoods along the observation grid.

Usage

make_functional_grouping_function(
  x,
  association = NULL,
  method = c("threshold", "community"),
  association_method = c("correlation", "neighborhood", "hybrid", "interval"),
  within_blocks = TRUE,
  bandwidth = NULL,
  interval_groups = NULL,
  width = NULL,
  step = width,
  decay = 1
)

Arguments

x

Any input accepted by as_functional_matrix().

association

Optional square association matrix. When omitted, the correlation matrix supplied by SelectBoost is reused after applying the FDA-specific masks.

method

Grouping strategy. "threshold" wraps SelectBoost::group_func_1() and "community" wraps SelectBoost::group_func_2().

association_method

Association structure passed to functional_association().

within_blocks

Should groups be restricted to features coming from the same functional block?

bandwidth

Optional maximum within-block lag retained in groups.

interval_groups, width, step, decay

Additional arguments passed to functional_association() when using region-aware associations.

Value

A function with signature ⁠(absXcor, c0)⁠ compatible with SelectBoost.

Smooth Trajectory Functional Example

Description

Simulated smooth trajectories used to demonstrate spline-basis and FPCA preprocessing from raw curves.

Usage

motion_example

Format

A list with four components:

grid

Numeric vector of observation times.

response

Numeric response vector.

predictors

Named list of functional predictor matrices.

scalar_covariates

Data frame with scalar covariates.

Source

Simulated for package examples.

Plain SelectBoost Baseline for Functional Predictors

Description

Runs SelectBoost directly on the flattened predictor matrix without the FDA-specific grouping heuristics used by selectboost_fda(). This is useful as a benchmark against the FDA-aware variant.

Usage

plain_selectboost(
  x,
  y = NULL,
  mode = c("fast", "auto"),
  selector = "msgps",
  selector_fun = NULL,
  selector_args = list(),
  groups = NULL,
  family = c("gaussian", "binomial"),
  association = NULL,
  group_method = c("threshold", "community"),
  ...
)

Arguments

x

Any input accepted by as_functional_matrix(), or an fda_design object.

y

Response vector. Leave as NULL when x is an fda_design.

mode

"fast" for a fixed c0 grid or "auto" for the adaptive version.

selector

Base selector used inside SelectBoost. Choose from "msgps", "lasso", "group_lasso", "sparse_group_lasso", the backend-specific aliases "glmnet", "grpreg", "sgl", or provide a custom function.

selector_fun

Optional custom base selector. It must return a coefficient vector of length p.

selector_args

Optional named list of arguments forwarded to the base selector.

groups

Optional feature groups used by grouped base selectors such as "grpreg". Defaults to block-level groups for list inputs.

family

Model family passed to built-in selectors.

association

Optional absolute association matrix used directly by the raw SelectBoost grouping function.

group_method

Functional grouping backend: threshold-based or community-based.

...

Additional arguments passed to SelectBoost::fastboost() or SelectBoost::autoboost().

Value

An object of class plain_selectboost_result.

Plot FDA Selection Results

Description

Plots feature-, group-, interval-, and basis-level summaries derived from selection_map(). The available views depend on the fitted object:

Usage

## S3 method for class 'fda_stability_selection'
plot(
  x,
  type = c("feature", "group", "interval", "basis"),
  value = c("group", "mean", "max"),
  facet = c("none", "predictor"),
  palette = selection_palette(),
  show_legend = TRUE,
  legend_title = NULL,
  legend_n_ticks = 3L,
  legend_digits = 2L,
  legend_cex = 0.75,
  cutoff = x$cutoff,
  ...
)

## S3 method for class 'selectboost_fda_result'
plot(
  x,
  type = c("feature", "group", "basis"),
  value = c("max", "mean"),
  palette = selection_palette(),
  show_legend = TRUE,
  legend_title = NULL,
  legend_n_ticks = 3L,
  legend_digits = 2L,
  legend_cex = 0.75,
  c0 = NULL,
  ...
)

Arguments

x

An object returned by stability_selection_fda(), interval_stability_selection(), fit_stability(), selectboost_fda(), or fit_selectboost().

type

Summary view to plot. Stability-selection fits support "feature", "group", "interval", and "basis". SelectBoost fits support "feature", "group", and "basis".

value

Quantity summarized in group, interval, and basis views. Stability-selection fits accept "group", "mean", and "max". SelectBoost fits accept "mean" and "max".

facet

Faceting mode for interval heatmaps. Currently only type = "interval" uses this argument.

palette

Vector of colors used for heatmaps.

show_legend

Logical; should heatmap views draw a legend?

legend_title

Optional legend title for heatmap views. By default an informative title is chosen from type and value.

legend_n_ticks

Approximate number of tick marks used in the heatmap legend.

legend_digits

Number of significant digits used for heatmap legend labels.

legend_cex

Character expansion used for heatmap legend text.

cutoff

Stability threshold. Only used for fda_stability_selection objects.

...

Additional graphical parameters passed to bar-plot-based views.

c0

Optional SelectBoost correlation threshold. When omitted, SelectBoost heatmaps are drawn across all available c0 values.

Details

Heatmap-based views are used for interval summaries and for SelectBoost summaries over multiple c0 values. Bar-plot views are used otherwise.

Value

Invisibly returns the helper output used to build the plot.

See Also

selection_map()

Run a Targeted Sensitivity Study for FDA-SelectBoost

Description

Repeats the FDA benchmark over a grid of simulation settings and a grid of FDA-aware SelectBoost settings. This is intended to answer the specific benchmark question of when selectboost_fda() improves on plain SelectBoost.

Usage

run_selectboost_sensitivity_study(
  n_rep = 10L,
  simulate_grid = expand.grid(scenario = c("localized_dense", "confounded_blocks"),
    confounding_strength = c(0.4, 0.9), active_region_scale = c(1, 0.7),
    local_correlation = c(0, 2), stringsAsFactors = FALSE),
  selectboost_grid = expand.grid(association_method = c("correlation", "neighborhood",
    "hybrid", "interval"), bandwidth = c(NA, 4, 8), stringsAsFactors = FALSE),
  simulate_args = list(),
  benchmark_args = list(),
  seed = NULL,
  keep_results = FALSE
)

Arguments

n_rep

Number of replications per setting combination.

simulate_grid

Data frame of simulation-setting combinations. Columns are merged into simulate_args and can include scenario, confounding_strength, active_region_scale, and local_correlation.

selectboost_grid

Data frame of selectboost_fda() setting combinations. Columns are merged into benchmark_args$selectboost_args and can include association_method, bandwidth, width, or step.

simulate_args

Named list forwarded to simulate_fda_scenario().

benchmark_args

Named list forwarded to benchmark_selection_methods(). When omitted, the study compares FDA-aware SelectBoost, plain SelectBoost, and grouped stability selection.

seed

Optional seed used to derive deterministic per-replication and per-setting seeds.

keep_results

Should the individual benchmark objects be returned?

Value

An object inheriting from fda_selectboost_sensitivity_study and fda_simulation_study.

Examples

grid <- data.frame(
  scenario = "confounded_blocks",
  confounding_strength = 0.9,
  active_region_scale = 0.7,
  local_correlation = 2,
  stringsAsFactors = FALSE
)
methods <- data.frame(
  association_method = c("correlation", "hybrid"),
  bandwidth = c(NA, 4),
  stringsAsFactors = FALSE
)
study <- run_selectboost_sensitivity_study(
  n_rep = 1,
  simulate_grid = grid,
  selectboost_grid = methods,
  simulate_args = list(n = 24, grid_length = 16),
  benchmark_args = list(
    methods = c("selectboost", "plain_selectboost"),
    levels = "feature",
    selectboost_args = list(B = 3, steps.seq = 0.5, c0lim = FALSE),
    plain_selectboost_args = list(B = 3, steps.seq = 0.5, c0lim = FALSE)
  ),
  seed = 1
)
summarise_benchmark_advantage(
  study,
  target = "selectboost",
  reference = "plain_selectboost",
  level = "feature"
)

Run a Repeated FDA Simulation Study

Description

Repeats simulate_fda_scenario() and benchmark_selection_methods() over multiple replications and aggregates the resulting recovery metrics.

Usage

run_simulation_study(
  n_rep = 10L,
  simulate_args = list(),
  benchmark_args = list(),
  seed = NULL,
  keep_results = FALSE
)

Arguments

n_rep

Number of simulation replications.

simulate_args

Named list forwarded to simulate_fda_scenario().

benchmark_args

Named list forwarded to benchmark_selection_methods().

seed

Optional seed used to derive deterministic per-replication seeds.

keep_results

Should the individual benchmark objects be returned?

Value

An object of class fda_simulation_study.

FDA-Oriented SelectBoost Wrapper

Description

Wraps SelectBoost::fastboost() or SelectBoost::autoboost() while adding FDA-specific structure through block-aware and region-aware grouping.

Usage

selectboost_fda(
  x,
  y = NULL,
  mode = c("fast", "auto"),
  selector = "msgps",
  selector_fun = NULL,
  selector_args = list(),
  groups = NULL,
  family = c("gaussian", "binomial"),
  association = NULL,
  group_method = c("threshold", "community"),
  association_method = c("correlation", "neighborhood", "hybrid", "interval"),
  within_blocks = TRUE,
  bandwidth = NULL,
  interval_groups = NULL,
  width = NULL,
  step = width,
  decay = 1,
  ...
)

Arguments

x

Any input accepted by as_functional_matrix(), or an fda_design object.

y

Response vector. Leave as NULL when x is an fda_design.

mode

"fast" for a fixed c0 grid or "auto" for the adaptive version.

selector

Base selector used inside SelectBoost. Choose from "msgps", "lasso", "group_lasso", "sparse_group_lasso", the backend-specific aliases "glmnet", "grpreg", "sgl", or provide a custom function.

selector_fun

Optional custom base selector. It must return a coefficient vector of length p.

selector_args

Optional named list of arguments forwarded to the base selector.

groups

Optional feature groups used by grouped base selectors such as "grpreg". Defaults to block-level groups for list inputs.

family

Model family passed to built-in selectors.

association

Optional custom association matrix used to define FDA-aware groups.

group_method

Functional grouping backend: threshold-based or community-based.

association_method

Association structure used to build FDA-aware groups.

within_blocks

Should SelectBoost groups stay within functional blocks?

bandwidth

Optional maximum within-block lag retained in groups.

interval_groups, width, step, decay

Additional arguments passed to make_functional_grouping_function().

...

Additional arguments passed to SelectBoost::fastboost() or SelectBoost::autoboost().

Value

An object of class selectboost_fda_result.

Extract Selected Features or Groups

Description

Returns the selected rows from selection_map() for stability-selection or SelectBoost fits.

Usage

selected(x, ...)

Arguments

x

A fitted selection object.

...

Additional arguments passed to the relevant method.

Value

A data frame.

Feature-Level Selection Map

Description

Returns a feature map augmented with selection summaries from a fit object.

Usage

selection_map(x, level = c("feature", "group", "basis"), ...)

Arguments

x

An fda_design, fda_stability_selection, or selectboost_fda_result object.

level

Summary level. "feature" returns one row per coefficient, "group" returns one row per stability/interval group, and "basis" returns one row per basis-expanded predictor.

...

Additional arguments passed to the relevant method.

Value

A data frame.

Simulate an FDA Benchmark Scenario

Description

Generates raw functional predictors, scalar covariates, a response, and the mapped ground truth for the transformed design matrix.

Usage

simulate_fda_scenario(
  n = 80L,
  grid_length = 60L,
  family = c("gaussian", "binomial"),
  representation = c("grid", "basis", "fpca"),
  transforms = NULL,
  basis_df = 7L,
  n_components = 5L,
  scenario = c("localized_dense", "distributed_smooth", "confounded_blocks"),
  confounding_strength = NULL,
  active_region_scale = 1,
  local_correlation = 0,
  include_scalar = TRUE,
  noise_sd = 0.4,
  seed = NULL
)

Arguments

n

Number of observations.

grid_length

Number of grid points per functional predictor.

family

Model family used to generate the response.

representation

Representation used when building the returned fda_design(): "grid" keeps the raw curves, "basis" applies a spline-basis transform, and "fpca" applies FPCA scores.

transforms

Optional transform list passed to fda_design(). When omitted, a sensible default is chosen from representation.

basis_df

Degrees of freedom used when representation = "basis".

n_components

Number of FPCA components used when representation = "fpca".

scenario

Benchmark scenario. "localized_dense" emphasizes narrow active regions under strong local correlation, "distributed_smooth" spreads the effect over broader smooth regions, and "confounded_blocks" adds stronger nuisance structure near the active block.

confounding_strength

Strength of cross-block confounding injected into the nuisance curve. Higher values make plain SelectBoost less able to separate true local signals from correlated nuisance structure.

active_region_scale

Positive multiplier applied to the width of the active regions. Values below 1 create narrower active regions.

local_correlation

Non-negative smoothing parameter applied to the simulated curves. Larger values increase local correlation along the grid.

include_scalar

Should scalar covariates be included in the design and truth object?

noise_sd

Observation noise level.

seed

Optional random seed.

Value

An object of class fda_simulation_data.

Examples

sim <- simulate_fda_scenario(n = 24, grid_length = 16, seed = 1)
sim
head(sim$truth$active_features)

Spectroscopy-Style Functional Example

Description

Simulated dense spectra with one signal block, one nuisance block, and two scalar covariates. The response is continuous and depends on localized regions of the signal spectrum plus the scalar covariates.

Usage

spectra_example

Format

A list with four components:

grid

Numeric vector of wavelength locations.

response

Numeric response vector.

predictors

Named list of functional predictor matrices.

scalar_covariates

Data frame with scalar covariates.

Source

Simulated for package examples.

Grouped Stability Selection for Functional Predictors

Description

Repeatedly subsamples observations, refits a sparse base selector, and computes exact feature- and group-level selection frequencies. This is the generic FDA recipe for basis expansions, discretized curves, or FPCA scores.

Usage

stability_selection_fda(
  x,
  y = NULL,
  selector = "group_lasso",
  selector_fun = NULL,
  groups = NULL,
  family = c("gaussian", "binomial"),
  B = 100L,
  sample_fraction = 0.5,
  cutoff = 0.75,
  seed = NULL,
  keep_subsamples = FALSE,
  ...
)

Arguments

x

Any input accepted by as_functional_matrix(), or an fda_design object.

y

Response vector. Leave as NULL when x is an fda_design.

selector

Either "lasso", "group_lasso", "sparse_group_lasso", one of the backend-specific aliases ("glmnet", "grpreg", "sgl"), or a custom function.

selector_fun

Optional custom selector. It must accept X, y, groups, and family, and return either a coefficient vector or a logical selection vector of length p.

groups

Optional grouping structure. Defaults to block-level groups when x is supplied as a list, and otherwise to one group per feature.

family

Model family passed to the built-in selectors.

B

Number of subsampling replicates.

sample_fraction

Fraction of observations drawn without replacement in each subsample.

cutoff

Stability threshold used to define selected_features and selected_groups.

seed

Optional random seed.

keep_subsamples

Should the sampled row indices be returned?

...

Additional arguments forwarded to the built-in or custom selector.

Value

An object of class fda_stability_selection.

Suggest a c0 Grid for FDA-SelectBoost

Description

Builds a data-driven c0 grid from an FDA-aware association matrix.

Usage

suggest_c0_grid(
  x,
  n = 5L,
  method = c("quantile", "linear"),
  association_method = c("correlation", "neighborhood", "hybrid", "interval"),
  within_blocks = TRUE,
  bandwidth = NULL,
  interval_groups = NULL,
  width = NULL,
  step = width,
  decay = 1
)

Arguments

x

Any input accepted by as_functional_matrix().

n

Number of grid values to return.

method

Grid construction rule: "quantile" or "linear".

association_method

Association structure passed to functional_association().

within_blocks, bandwidth, interval_groups, width, step, decay

Passed to functional_association().

Value

A decreasing numeric vector of c0 values.

Summarize the Advantage of FDA-SelectBoost Over Baselines

Description

Computes the per-scenario and per-level gain of a target method over one or more reference methods. This is intended to make the benchmark story explicit when comparing FDA-aware SelectBoost to existing baselines.

Usage

summarise_benchmark_advantage(
  x,
  target = "selectboost",
  reference = c("plain_selectboost", "stability"),
  level = c("feature", "group", "basis"),
  metric = "f1",
  optimize = c("max", "min"),
  select_c0 = c("best", "all")
)

Arguments

x

An fda_benchmark or fda_simulation_study object.

target

Method whose gain should be assessed.

reference

One or more baseline methods.

level

Evaluation level.

metric

Metric used both for best-c0 selection and for the reported gains.

optimize

Should larger or smaller values of metric be preferred?

select_c0

Keep all c0 rows or only the best one per method and replicate.

Value

A data frame.

Summarize Benchmark Performance by Method

Description

Collapses raw benchmark rows into method-level performance summaries, with an option to retain only the best c0 per method and replication.

Usage

summarise_benchmark_performance(
  x,
  level = c("feature", "group", "basis"),
  metric = "f1",
  optimize = c("max", "min"),
  select_c0 = c("best", "all")
)

Arguments

x

An fda_benchmark or fda_simulation_study object.

level

Evaluation level.

metric

Metric used to pick the best c0 when select_c0 = "best".

optimize

Should larger or smaller values of metric be preferred?

select_c0

Keep all c0 rows or only the best one per method and replicate.

Value

A data frame.