Package {AutoMLR}

Type: Package
Title: Automated Multi-Outcome Machine Learning Combination Models
Version: 1.0.0
Description: Provides automated machine learning workflows for survival analysis, binary classification, continuous outcomes, and ordinal outcomes. The package trains and combines model variants across user-supplied multi-cohort data, evaluates survival models by leave-one-out cross-validation using Harrell's concordance index, binary models by leave-one-out cross-validation using receiver operating characteristic area under the curve, continuous models by out-of-fold root mean squared error and R-squared, and ordinal models by out-of-fold quadratic weighted kappa. It renders reproducible reports in Hypertext Markup Language (HTML) with figures and diagnostics. The survival workflow supports penalized and tree-based Cox proportional hazards models, stepwise Cox models, partial least squares regression for Cox models, supervised principal components, gradient boosting machine Cox models, survival support vector machines (survival-SVM), random survival forests, and optional 'CoxBoost'. The binary workflow supports penalized logistic regression, logistic baselines, gradient boosting machines, random forests, principal component analysis (PCA) logistic regression, and Gaussian naive Bayes variants. Continuous and ordinal workflows reuse an 18-variant regression registry with penalized, linear, boosted, forest, PCA, and baseline families. The optional 'CoxBoost' model is enabled when the suggested 'CoxBoost' package is installed; it is used conditionally and is not a strong dependency. Optional model backends are checked at run time so missing backend packages skip only the affected model variants rather than blocking installation of the whole package. Methods build on Friedman et al. (2010) <doi:10.18637/jss.v033.i01>, Bair and Tibshirani (2004) <doi:10.1371/journal.pbio.0020108>, Ishwaran et al. (2008) <doi:10.1214/08-AOAS169>, Blanche et al. (2013) <doi:10.1002/sim.5958>, and Binder and Schumacher (2008) <doi:10.1186/1471-2105-9-14>.
License: MIT + file LICENSE
Encoding: UTF-8
Depends: R (≥ 4.1)
Imports: survival, graphics, grDevices, parallel, stats, utils
Suggests: CoxBoost, digest, future, future.apply, glmnet, gbm, log4r, plsRcox, quadprog, randomForestSRC, superpc, survivalsvm, testthat (≥ 3.0.0), timeROC
Config/testthat/edition: 3
RoxygenNote: 7.3.3
NeedsCompilation: no
Packaged: 2026-05-19 16:21:46 UTC; abc
Author: Peng Luo [aut, cre]
Maintainer: Peng Luo <luopeng@smu.edu.cn>
Repository: CRAN
Date/Publication: 2026-06-07 18:50:19 UTC

Extract modeling matrices from prepared binary input.

Description

Converts an object returned by prepare_binary_cohort_input() into the X, y, and cohort-label objects used by lower-level binary functions such as evaluate_binary_algorithm_loocv(), evaluate_binary_algorithms_loocv(), and evaluate_binary_combinations().

Usage

automlr_input_to_binary_xy(input)

Arguments

input

An object returned by prepare_binary_cohort_input().

Value

A list with components X (numeric feature matrix), y (0/1 outcome), stability_groups (cohort labels aligned to rows), cohort (alias of stability_groups), and data (combined modeling data frame).

Extract modeling matrices from prepared continuous input.

Description

Converts an object returned by prepare_continuous_cohort_input() into the X, y, and cohort-label objects used by lower-level continuous-outcome functions.

Usage

automlr_input_to_continuous_xy(input)

Arguments

input

An object returned by prepare_continuous_cohort_input().

Value

A list with components X (numeric feature matrix), y (numeric outcome), stability_groups (cohort labels aligned to rows), cohort (alias of stability_groups), and data (combined modeling data frame).

Extract modeling matrices from prepared ordinal input.

Description

Converts an object returned by prepare_ordinal_cohort_input() into the X, y, and cohort-label objects used by lower-level ordinal-outcome functions.

Usage

automlr_input_to_ordinal_xy(input)

Arguments

input

An object returned by prepare_ordinal_cohort_input().

Value

A list with components X (numeric feature matrix), y (integer ordinal outcome codes), stability_groups (cohort labels aligned to rows), cohort (alias of stability_groups), and data (combined modeling data frame).

Extract modeling matrices from prepared survival input.

Description

Converts an object returned by prepare_cohort_input() into the X, y, and cohort-label objects used by lower-level survival functions such as evaluate_algorithm_loocv(), evaluate_algorithms_loocv(), and evaluate_surv_combinations().

Usage

automlr_input_to_surv_xy(input)

Arguments

input

An object returned by prepare_cohort_input().

Value

A list with components:

X

Numeric feature matrix restricted to shared features.

y

A survival::Surv outcome object.

stability_groups

Cohort labels aligned to rows of X.

cohort

Alias of stability_groups.

data

The combined modeling data frame.

Default parameters for AutoMLR survival pipeline.

Description

Returns a flat list of knobs consumed by data preparation, feature screening, model fitting, and evaluation.

Usage

automlr_parameters(
  seed = 123L,
  screen_by_univariate_cox = TRUE,
  univariate_cox_p_cutoff = 0.05,
  screen_by_variance = TRUE,
  variance_quantile_cutoff = 0.1,
  loocv = TRUE,
  min_cindex_accept = 0.6,
  auto_min_cindex = FALSE,
  auto_quantile = 0.5,
  eval_times = c(365, 1095, 1825),
  time_unit = c("auto", "days", "months", "years"),
  algorithms = NULL,
  surv_svm_resampling = c("kfold", "loocv"),
  surv_svm_k_folds = 5L,
  n_cores = 1L,
  stability_resamples = 0L,
  stability_fraction = 0.8,
  stability_weight = 0.1,
  verbose = TRUE
)

Arguments

seed

Base random seed.

screen_by_univariate_cox

Logical, run univariate Cox p-value screen.

univariate_cox_p_cutoff

P-value cutoff for the univariate Cox screen.

screen_by_variance

Logical, drop low-variance features.

variance_quantile_cutoff

Drop features with variance below this quantile of all feature variances (e.g. 0.1 = drop bottom 10%).

loocv

Logical, use LOOCV during model selection (V1 evaluation protocol; paper 1 recipe).

min_cindex_accept

Minimum LOOCV C-index to accept a model.

auto_min_cindex

Logical. If TRUE and fit_surv_ensemble() uses strategy = "threshold", replace min_cindex_accept with a recommended cutoff computed from candidate-model LOOCV C-index values.

auto_quantile

Quantile used when an automatic threshold is requested. 0.50 uses the median candidate metric; larger values are more selective.

eval_times

Vector of time points (same unit as time column) at which to compute time-dependent AUC. Default ~1-, 3-, 5-year style.

time_unit

Time unit for survival time and eval_times, one of "auto", "days", "months", or "years". "auto" infers the unit from the observed follow-up scale for plot labels only.

algorithms

Character vector of algorithm keys to run; default = all 10 in list_surv_algorithms().

surv_svm_resampling

Resampling method for survival support vector machine candidates. The default "kfold" avoids known single-row prediction failures from survivalsvm during LOOCV. Set "loocv" to use leave-one-out evaluation for this model as well.

surv_svm_k_folds

Number of folds used when surv_svm_resampling = "kfold".

n_cores

Integer, workers for parallel execution. 1 = sequential.

stability_resamples

Integer, number of repeated subsamples used for optional stability diagnostics. 0 disables resampling.

stability_fraction

Fraction of samples used in each stability subsample.

stability_weight

Non-negative penalty multiplier used only when a stability-weighted ranking or weight method is explicitly requested.

verbose

Logical, print progress messages.

Value

A named list.

ROC AUC for binary outcomes.

Description

ROC AUC for binary outcomes.

Usage

binary_auc(y, prob)

Arguments

y

Binary 0/1 outcome.

prob

Predicted probability for the positive class.

Value

Scalar ROC AUC, or NA_real_ when not estimable.

Precision-recall AUC for binary outcomes.

Description

Precision-recall AUC for binary outcomes.

Usage

binary_pr_auc(y, prob)

Arguments

y

Binary 0/1 outcome.

prob

Predicted probability for the positive class.

Value

Scalar PR-AUC, or NA_real_ when not estimable.

Default parameters for AutoMLR binary-classification workflows.

Description

The binary workflow mirrors the survival workflow, but uses cross-validated ROC AUC as the primary selection metric and keeps PR-AUC, threshold metrics, calibration, and cohort stability as diagnostics.

Usage

binarymlr_parameters(
  seed = 123L,
  algorithms = NULL,
  loocv = TRUE,
  resampling = NULL,
  k_folds = 5L,
  repeats = 1L,
  min_auc_accept = 0.6,
  auto_min_auc = FALSE,
  auto_quantile = 0.5,
  positive_class = 1,
  threshold_methods = c("youden", "fixed_0.5"),
  missing_fraction_cutoff = 0.2,
  screen_by_variance = TRUE,
  variance_quantile_cutoff = 0,
  standardize_features = FALSE,
  n_cores = 1L,
  stability_resamples = 0L,
  stability_fraction = 0.8,
  stability_weight = 0.1,
  verbose = TRUE
)

Arguments

seed

Base random seed.

algorithms

Character vector of binary algorithm keys. Defaults to all entries returned by list_binary_algorithms().

loocv

Logical, use leave-one-out cross-validation.

resampling

Resampling scheme: "loocv", "kfold", or "repeated_kfold".

k_folds

Number of folds when resampling uses k-fold CV.

repeats

Number of repeats for repeated k-fold CV.

min_auc_accept

Minimum AUC accepted by threshold-style selection.

auto_min_auc

Logical. If TRUE and fit_binary_ensemble() uses strategy = "threshold", replace min_auc_accept with a recommended cutoff computed from candidate-model AUC values.

auto_quantile

Quantile used when an automatic threshold is requested. 0.50 uses the median candidate metric; larger values are more selective.

positive_class

Positive-class label used during data preparation.

threshold_methods

Threshold summaries to export. Supported values are "youden" and "fixed_0.5".

missing_fraction_cutoff

Drop features with missing fraction above this cutoff before modeling.

screen_by_variance

Logical, drop zero / low-variance features.

variance_quantile_cutoff

Optional lower variance quantile to drop.

standardize_features

Logical, center and scale features before modeling.

n_cores

Integer, number of fold workers. 1 is sequential.

stability_resamples

Number of optional stability subsamples.

stability_fraction

Fraction of samples in each stability subsample.

stability_weight

Penalty multiplier for stability-weighted ranking.

verbose

Logical, print progress.

Value

A named list.

Check optional AutoMLR model backends and feature dependencies.

Description

AutoMLR keeps heavyweight model engines in Suggests so the package can be installed even when some optional modelling backends are unavailable. This helper reports which model variants and optional features are currently available in the user's R library and which packages would be needed to enable the skipped pieces.

Usage

check_automlr_dependencies(workflow = "all")

Arguments

workflow

Character vector. Use "all" to check survival, binary, continuous, and ordinal workflows, or select any subset of "survival", "binary", "continuous", and "ordinal".

Value

A list of class "automlr_dependency_report" with two data frames: algorithms, containing one row per algorithm registry entry, and optional_features, containing non-model optional capabilities such as logging, parallel execution, and time-dependent ROC diagnostics.

Correlation between observed and predicted continuous outcomes.

Description

Correlation between observed and predicted continuous outcomes.

Usage

continuous_cor(y, pred, method = c("pearson", "spearman"))

Arguments

y

Observed numeric outcome.

pred

Predicted numeric outcome.

method

Correlation method.

Value

A numeric scalar.

Mean absolute error for continuous predictions.

Description

Mean absolute error for continuous predictions.

Usage

continuous_mae(y, pred)

Arguments

y

Observed numeric outcome.

pred

Predicted numeric outcome.

Value

A numeric scalar.

Coefficient of determination for continuous predictions.

Description

Coefficient of determination for continuous predictions.

Usage

continuous_r2(y, pred)

Arguments

y

Observed numeric outcome.

pred

Predicted numeric outcome.

Value

A numeric scalar.

Root mean squared error for continuous predictions.

Description

Root mean squared error for continuous predictions.

Usage

continuous_rmse(y, pred)

Arguments

y

Observed numeric outcome.

pred

Predicted numeric outcome.

Value

A numeric scalar.

Default parameters for AutoMLR continuous-outcome workflows.

Description

Default parameters for AutoMLR continuous-outcome workflows.

Usage

continuousmlr_parameters(
  seed = 123L,
  algorithms = NULL,
  resampling = "loocv",
  k_folds = 5L,
  repeats = 1L,
  min_r2_accept = 0,
  auto_min_r2 = FALSE,
  auto_quantile = 0.5,
  missing_fraction_cutoff = 0.2,
  screen_by_variance = TRUE,
  variance_quantile_cutoff = 0,
  standardize_features = FALSE,
  n_cores = 1L,
  stability_resamples = 0L,
  stability_fraction = 0.8,
  stability_weight = 0.1,
  verbose = TRUE
)

Arguments

seed

Base random seed.

algorithms

Character vector of continuous algorithm keys.

resampling

Resampling scheme: "loocv", "kfold", or "repeated_kfold".

k_folds

Number of folds for k-fold CV.

repeats

Number of repeats for repeated k-fold CV.

min_r2_accept

Minimum R-squared accepted by threshold-style selection.

auto_min_r2

Logical. If TRUE and fit_continuous_ensemble() uses strategy = "threshold", replace min_r2_accept with a recommended cutoff computed from candidate-model R-squared values.

auto_quantile

Quantile used when an automatic threshold is requested. 0.50 uses the median candidate metric; larger values are more selective.

missing_fraction_cutoff

Drop features above this missing fraction.

screen_by_variance

Logical, drop zero / low-variance features.

variance_quantile_cutoff

Optional lower variance quantile to drop.

standardize_features

Logical, center and scale features.

n_cores

Integer, number of fold workers.

stability_resamples

Number of optional stability subsamples.

stability_fraction

Fraction of samples in each stability subsample.

stability_weight

Penalty multiplier for stability-aware ranking.

verbose

Logical, print progress.

Value

A named list.

Count binary model combinations without fitting.

Description

Count binary model combinations without fitting.

Usage

count_binary_combinations(
  params = binarymlr_parameters(),
  algorithms = params$algorithms,
  min_size = 1L,
  max_size = 2L,
  allow_same_algorithm = FALSE
)

Arguments

params

Output of binarymlr_parameters().

algorithms

Binary algorithm keys.

min_size

Minimum combination size.

max_size

Maximum combination size.

allow_same_algorithm

Logical, allow same base algorithm variants in one combination.

Value

A list with candidate and combination counts.

Count continuous model combinations without fitting.

Description

Count continuous model combinations without fitting.

Usage

count_continuous_combinations(
  params = continuousmlr_parameters(),
  algorithms = params$algorithms,
  min_size = 1L,
  max_size = 2L,
  allow_same_algorithm = FALSE
)

Arguments

params

Output of continuousmlr_parameters().

algorithms

Continuous algorithm keys.

min_size

Minimum combination size.

max_size

Maximum combination size.

allow_same_algorithm

Logical.

Value

A list with candidate and combination counts.

Count ordinal model combinations without fitting.

Description

Count ordinal model combinations without fitting.

Usage

count_ordinal_combinations(
  params = ordinalmlr_parameters(),
  algorithms = params$algorithms,
  min_size = 1L,
  max_size = 2L,
  allow_same_algorithm = FALSE
)

Arguments

params

Output of ordinalmlr_parameters().

algorithms

Ordinal algorithm keys.

min_size

Minimum combination size.

max_size

Maximum combination size.

allow_same_algorithm

Logical.

Value

A list with candidate and combination counts.

Count model combinations without fitting models.

Description

Count model combinations without fitting models.

Usage

count_surv_combinations(
  params = automlr_parameters(),
  algorithms = params$algorithms,
  min_size = 2L,
  max_size = 2L,
  allow_same_algorithm = FALSE
)

Arguments

params

Output of automlr_parameters().

algorithms

Character vector of registry keys.

min_size

Minimum combination size.

max_size

Maximum combination size.

allow_same_algorithm

Logical, allow two variants from the same base algorithm to appear in one combination.

Value

A list with candidate and combination counts.

Disable AutoMLR auto logging.

Description

Disable AutoMLR auto logging.

Usage

disable_auto_logging()

Value

Invisibly returns TRUE when logging was disabled and FALSE when logging was already inactive.

Run LOOCV for a named algorithm in the registry.

Description

Convenience wrapper: looks up algo_key in get_surv_registry(), takes the first row of grid(params) as the hyperparameters, and calls loocv_cindex().

Usage

evaluate_algorithm_loocv(
  algo_key,
  X,
  y,
  params = automlr_parameters(),
  hparam = NULL,
  candidate_key = NULL,
  verbose = FALSE
)

Arguments

algo_key

Character, one of list_surv_algorithms().

X

Feature matrix.

y

survival::Surv object.

params

Output of automlr_parameters(); used to pick the grid row.

hparam

Optional named list of hyperparameters. If NULL, the first row of the registry grid is used.

candidate_key

Optional identifier for this algorithm + hyperparameter variant.

verbose

Logical.

Value

A list — the output of loocv_cindex() plus algo_key, algo_label, candidate_key, and hparam used.

Evaluate multiple survival model variants by LOOCV C-index.

Description

Evaluate multiple survival model variants by LOOCV C-index.

Usage

evaluate_algorithms_loocv(
  X,
  y,
  params = automlr_parameters(),
  algorithms = params$algorithms,
  stability_groups = NULL,
  stability_resamples = params$stability_resamples %||% 0L,
  stability_fraction = params$stability_fraction %||% 0.8,
  verbose = params$verbose
)

Arguments

X

Feature matrix, or an "automlr_input" object returned by prepare_cohort_input().

y

survival::Surv object. Leave NULL when X is an "automlr_input" object.

params

Output of automlr_parameters().

algorithms

Character vector of registry keys to expand and evaluate.

stability_groups

Optional group/queue labels used to compute per-candidate stability diagnostics from out-of-fold risks.

stability_resamples

Number of repeated subsamples for stability diagnostics.

stability_fraction

Fraction of samples in each stability subsample.

verbose

Logical.

Value

A list of class "automlr_loocv_set" with a summary table and raw per-candidate results.

Evaluate one binary algorithm by LOOCV AUC.

Description

Evaluate one binary algorithm by LOOCV AUC.

Usage

evaluate_binary_algorithm_loocv(
  algo_key,
  X,
  y,
  params = binarymlr_parameters(),
  hparam = NULL,
  candidate_key = NULL,
  verbose = FALSE,
  resampling_plan = NULL
)

Arguments

algo_key

Algorithm key.

X

Feature matrix.

y

Binary 0/1 outcome.

params

Output of binarymlr_parameters().

hparam

Optional hyperparameters.

candidate_key

Optional variant identifier.

verbose

Logical.

resampling_plan

Optional internal resampling plan.

Value

A list with LOOCV results and metadata.

Evaluate binary model variants by LOOCV AUC.

Description

Evaluate binary model variants by LOOCV AUC.

Usage

evaluate_binary_algorithms_loocv(
  X,
  y,
  params = binarymlr_parameters(),
  algorithms = params$algorithms,
  stability_groups = NULL,
  stability_resamples = params$stability_resamples %||% 0L,
  stability_fraction = params$stability_fraction %||% 0.8,
  verbose = params$verbose
)

Arguments

X

Feature matrix.

y

Binary 0/1 outcome.

params

Output of binarymlr_parameters().

algorithms

Algorithm keys.

stability_groups

Optional cohort labels.

stability_resamples

Optional stability resamples.

stability_fraction

Stability subsample fraction.

verbose

Logical.

Value

A list of class "automlr_binary_loocv_set".

Evaluate all-subset binary probability combinations.

Description

Evaluate all-subset binary probability combinations.

Usage

evaluate_binary_combinations(
  loocv_set,
  y,
  min_size = 1L,
  max_size = 2L,
  weight_method = c("auc", "equal", "auc_stability"),
  allow_same_algorithm = FALSE,
  max_failed_fraction = 0.2,
  min_prob_sd = 1e-08,
  stability_groups = NULL,
  stability_resamples = loocv_set$params$stability_resamples %||% 0L,
  stability_fraction = loocv_set$params$stability_fraction %||% 0.8,
  rank_by = c("auc", "stability_weighted"),
  stability_weight = loocv_set$params$stability_weight %||% 0.1,
  top_n = 50L
)

Arguments

loocv_set

Output of evaluate_binary_algorithms_loocv().

y

Binary 0/1 outcome.

min_size

Minimum member count.

max_size

Maximum member count.

weight_method

One of "auc", "equal", or "auc_stability".

allow_same_algorithm

Logical.

max_failed_fraction

Maximum LOOCV failure fraction.

min_prob_sd

Minimum probability standard deviation.

stability_groups

Optional cohort labels.

stability_resamples

Optional stability resamples.

stability_fraction

Stability subsample fraction.

rank_by

Ranking method.

stability_weight

Stability penalty multiplier.

top_n

Number of rows to keep.

Value

A list of class "automlr_binary_combination_set".

Evaluate one continuous algorithm by out-of-fold performance.

Description

Evaluate one continuous algorithm by out-of-fold performance.

Usage

evaluate_continuous_algorithm(
  algo_key,
  X,
  y,
  params = continuousmlr_parameters(),
  hparam = NULL,
  candidate_key = NULL,
  verbose = FALSE,
  resampling_plan = NULL
)

Arguments

algo_key

Algorithm key.

X

Feature matrix.

y

Numeric outcome.

params

Output of continuousmlr_parameters().

hparam

Optional hyperparameters.

candidate_key

Optional variant identifier.

verbose

Logical.

resampling_plan

Optional internal resampling plan.

Value

A list with OOF predictions and metrics.

Evaluate continuous model variants by out-of-fold performance.

Description

Evaluate continuous model variants by out-of-fold performance.

Usage

evaluate_continuous_algorithms(
  X,
  y,
  params = continuousmlr_parameters(),
  algorithms = params$algorithms,
  verbose = params$verbose
)

Arguments

X

Feature matrix.

y

Numeric outcome.

params

Output of continuousmlr_parameters().

algorithms

Algorithm keys.

verbose

Logical.

Value

A list of class "automlr_continuous_resample_set".

Evaluate all-subset continuous prediction combinations.

Description

Evaluate all-subset continuous prediction combinations.

Usage

evaluate_continuous_combinations(
  resample_set,
  y,
  min_size = 1L,
  max_size = 2L,
  weight_method = c("inverse_rmse", "equal", "r2"),
  allow_same_algorithm = FALSE,
  max_failed_fraction = 0.2,
  min_pred_sd = 1e-08,
  rank_by = c("rmse", "r2"),
  top_n = 50L
)

Arguments

resample_set

Output of evaluate_continuous_algorithms().

y

Numeric outcome.

min_size

Minimum member count.

max_size

Maximum member count.

weight_method

One of "inverse_rmse", "equal", or "r2".

allow_same_algorithm

Logical.

max_failed_fraction

Maximum sample prediction failure fraction.

min_pred_sd

Minimum prediction standard deviation.

rank_by

Ranking method.

top_n

Number of rows to keep.

Value

A list of class "automlr_continuous_combination_set".

Evaluate ordinal model variants by out-of-fold performance.

Description

Ordinal outcomes are encoded as ordered integer scores and fitted with the continuous-model registry; predictions are rounded back to ordered classes for QWK, accuracy, balanced accuracy, and class MAE.

Usage

evaluate_ordinal_algorithms(
  X,
  y,
  params = ordinalmlr_parameters(),
  algorithms = params$algorithms,
  verbose = params$verbose
)

Arguments

X

Feature matrix.

y

Ordinal positive integer outcome codes.

params

Output of ordinalmlr_parameters().

algorithms

Algorithm keys.

verbose

Logical.

Value

A list of class "automlr_ordinal_resample_set".

Evaluate all-subset ordinal score combinations.

Description

Evaluate all-subset ordinal score combinations.

Usage

evaluate_ordinal_combinations(
  resample_set,
  y,
  min_size = 1L,
  max_size = 2L,
  weight_method = c("qwk", "equal", "inverse_mae"),
  allow_same_algorithm = FALSE,
  rank_by = c("qwk", "class_mae"),
  top_n = 50L
)

Arguments

resample_set

Output of evaluate_ordinal_algorithms().

y

Ordinal integer outcome codes.

min_size

Minimum member count.

max_size

Maximum member count.

weight_method

One of "qwk", "equal", or "inverse_mae".

allow_same_algorithm

Logical.

rank_by

Ranking method.

top_n

Number of rows to keep.

Value

A list of class "automlr_ordinal_combination_set".

Evaluate all-subset survival model combinations.

Description

Builds combinations from the out-of-fold risk scores in evaluate_algorithms_loocv(). For each subset, member risks are standardized and averaged using equal weights or C-index-derived weights; the resulting combination is scored by Harrell's C-index.

Usage

evaluate_surv_combinations(
  loocv_set,
  y,
  min_size = 1L,
  max_size = 2L,
  weight_method = c("cindex", "equal", "cindex_stability"),
  allow_same_algorithm = FALSE,
  max_failed_fraction = 0.2,
  min_risk_sd = 1e-08,
  stability_groups = NULL,
  stability_resamples = loocv_set$params$stability_resamples %||% 0L,
  stability_fraction = loocv_set$params$stability_fraction %||% 0.8,
  rank_by = c("cindex", "stability_weighted"),
  stability_weight = loocv_set$params$stability_weight %||% 0.1,
  diagnostic_times = NULL,
  top_n = 50L
)

Arguments

loocv_set

Output of evaluate_algorithms_loocv().

y

A survival::Surv object used to score combined risks.

min_size

Minimum number of member model variants in a combination.

max_size

Maximum number of member model variants in a combination.

weight_method

One of "cindex", "equal", or "cindex_stability".

allow_same_algorithm

Logical, allow two variants from the same base algorithm to appear in one combination.

max_failed_fraction

Maximum allowed LOOCV fold failure fraction for a candidate to enter combinations.

min_risk_sd

Minimum standard deviation of LOOCV risk scores for a candidate to enter combinations.

stability_groups

Optional group/queue labels used for combination stability diagnostics.

stability_resamples

Number of repeated subsamples for stability diagnostics.

stability_fraction

Fraction of samples in each stability subsample.

rank_by

Ranking method. "cindex" keeps C-index as the primary selection criterion; "stability_weighted" subtracts a stability penalty.

stability_weight

Non-negative multiplier for the stability penalty when rank_by = "stability_weighted".

diagnostic_times

Optional time points for time-dependent AUC diagnostics. Requires the suggested timeROC package.

top_n

Number of top rows to keep in the returned summary.

Value

A list of class "automlr_combination_set".

Export binary AutoMLR results.

Description

Export binary AutoMLR results.

Usage

export_binary_results(
  object,
  output_dir = "automlr_binary_results",
  publication = TRUE,
  formats = c("pdf", "png"),
  dpi = 300L,
  top_n = 20L,
  overwrite = TRUE,
  summary_language = c("bilingual", "en", "zh")
)

Arguments

object

Object returned by fit_binary_ensemble().

output_dir

Output directory.

publication

Logical, write publication-style figures.

formats

Figure formats: "png" and/or "pdf".

dpi

PNG resolution.

top_n

Number of top rows.

overwrite

Logical.

summary_language

"bilingual", "en", or "zh".

Value

A list of exported paths.

Export continuous AutoMLR results.

Description

Export continuous AutoMLR results.

Usage

export_continuous_results(
  object,
  output_dir = "automlr_continuous_results",
  publication = TRUE,
  formats = c("pdf", "png"),
  dpi = 300L,
  top_n = 20L,
  overwrite = TRUE
)

Arguments

object

Object returned by fit_continuous_ensemble().

output_dir

Output directory.

publication

Logical, write publication-style figures.

formats

Figure formats.

dpi

PNG resolution.

top_n

Number of top rows.

overwrite

Logical.

Value

A list of exported paths.

Export extreme-screening tables and publication-style audit figures

Description

Writes the complete apparent C-index table, apparent top-N table, seed-search table, best rows, failure diagnostics, summary CSVs, a Markdown summary report, and a Morandi-toned audit figure set for an object returned by extreme_surv_screen().

Usage

export_extreme_screen_results(
  x,
  output_dir,
  formats = c("png", "pdf"),
  top_n = 10L,
  top_seed_rows = 30L,
  dpi = 300L,
  summary_language = c("bilingual", "en", "zh")
)

Arguments

x

An object returned by extreme_surv_screen().

output_dir

Directory to write tables, figures, and the RDS object.

formats

Figure formats. Any subset of "png" and "pdf".

top_n

Number of apparent top combinations to emphasize in figures.

top_seed_rows

Number of seed-search rows to show in ranked-row figures.

dpi

PNG resolution.

summary_language

Language used in summary_report.md, either "bilingual", "zh", or "en".

Value

A list with written table and figure paths.

Export ordinal AutoMLR results.

Description

Export ordinal AutoMLR results.

Usage

export_ordinal_results(
  object,
  output_dir = "automlr_ordinal_results",
  publication = TRUE,
  formats = c("pdf", "png"),
  dpi = 300L,
  top_n = 20L,
  overwrite = TRUE
)

Arguments

object

Object returned by fit_ordinal_ensemble().

output_dir

Output directory.

publication

Logical, write publication-style figures.

formats

Figure formats.

dpi

PNG resolution.

top_n

Number of top rows.

overwrite

Logical.

Value

A list of exported paths.

Export AutoMLR survival results as a reproducible result bundle.

Description

Creates a directory containing an HTML report, publication figures, diagnostic figures, CSV tables, fitted R objects, risk scores, optional timeROC diagnostics, cohort diagnostics, a deduplicated final publication figure set, and session metadata.

Usage

export_surv_results(
  object,
  output_dir = "automlr_results",
  publication = TRUE,
  formats = c("pdf", "png"),
  dpi = 300L,
  top_n = 20L,
  overwrite = TRUE,
  summary_language = c("bilingual", "en", "zh")
)

Arguments

object

An object returned by fit_surv_ensemble().

output_dir

Directory for the exported result bundle.

publication

Logical, create publication-style figures.

formats

Figure formats. Supported values are "pdf" and "png".

dpi

Resolution for PNG figures.

top_n

Number of top models / combinations to show in figures.

overwrite

Logical, overwrite report files where applicable.

summary_language

Language used in Markdown summaries, either "bilingual", "en", or "zh".

Value

A list with paths to exported files.

Extreme two-stage screening for survival model combinations

Description

Runs an optimistic "full data as train and validation" screen first, then searches random 70/30 train-validation splits only for the top combinations. The first stage estimates an apparent upper bound and should not be reported as external validation performance. Returned apparent-stage tables include performance_scope and performance_note columns carrying this warning.

Usage

extreme_surv_screen(
  X,
  y = NULL,
  params = automlr_parameters(),
  algorithms = params$algorithms,
  top_n = 5L,
  seeds = 1:500,
  train_fraction = 0.7,
  min_models = 1L,
  max_models = 2L,
  weight_method = c("cindex", "equal"),
  allow_same_algorithm = FALSE,
  min_risk_sd = 1e-08,
  rank_by = c("apparent_cindex", "mean_cohort_cindex"),
  stability_groups = NULL,
  n_cores = params$n_cores %||% 1L,
  verbose = params$verbose
)

Arguments

X

Feature matrix, or an "automlr_input" object returned by prepare_cohort_input().

y

survival::Surv object. Leave NULL when X is an "automlr_input" object.

params

Output of automlr_parameters().

algorithms

Character vector of registry keys.

top_n

Number of apparent-screen combinations carried into the seed search.

seeds

Integer vector of random seeds used for 70/30 split search.

train_fraction

Fraction of samples assigned to training in the seed search.

min_models

Minimum combination size in the apparent screen.

max_models

Maximum combination size in the apparent screen.

weight_method

One of "cindex" or "equal".

allow_same_algorithm

Logical, allow multiple variants from the same base algorithm in one combination.

min_risk_sd

Minimum apparent risk-score standard deviation required for a candidate to enter the apparent combination screen.

rank_by

"apparent_cindex" ranks by the pooled apparent C-index; "mean_cohort_cindex" ranks by cohort-average apparent C-index when cohort labels are available.

stability_groups

Optional group/cohort labels. Automatically taken from prepare_cohort_input() when X is an "automlr_input".

n_cores

Reserved for future seed-level parallel execution. The seed search currently runs sequentially for reproducible access to package internals.

verbose

Logical.

Value

A list of class "automlr_extreme_screen" with apparent-screen summaries, top combinations, seed-search results, and the best seed/model row by validation C-index. The notes component stores interpretation text for apparent performance and seed-search performance.

Fit a binary probability ensemble.

Description

Fit a binary probability ensemble.

Usage

fit_binary_ensemble(
  X,
  y = NULL,
  params = binarymlr_parameters(),
  algorithms = params$algorithms,
  min_auc = params$min_auc_accept,
  auto_min_auc = params$auto_min_auc %||% FALSE,
  auto_quantile = params$auto_quantile %||% 0.5,
  strategy = c("best_subset", "threshold"),
  min_models = 1L,
  max_models = 2L,
  weight_method = c("auc", "equal", "auc_stability"),
  allow_same_algorithm = FALSE,
  max_failed_fraction = 0.2,
  stability_groups = NULL,
  stability_resamples = params$stability_resamples %||% 0L,
  stability_fraction = params$stability_fraction %||% 0.8,
  rank_by = c("auc", "stability_weighted"),
  stability_weight = params$stability_weight %||% 0.1,
  threshold_method = "youden",
  verbose = params$verbose
)

Arguments

X

Feature matrix, or an "automlr_binary_input" object.

y

Binary 0/1 outcome. Leave NULL when X is an input object.

params

Output of binarymlr_parameters().

algorithms

Binary algorithm keys.

min_auc

Minimum AUC for threshold strategy.

auto_min_auc

Logical. If TRUE, compute min_auc from the candidate AUC distribution using auto_quantile.

auto_quantile

Quantile used for automatic threshold selection.

strategy

"best_subset" or "threshold".

min_models

Minimum combination size.

max_models

Maximum combination size.

weight_method

One of "auc", "equal", or "auc_stability".

allow_same_algorithm

Logical.

max_failed_fraction

Maximum fold-failure fraction.

stability_groups

Optional cohort labels.

stability_resamples

Stability resamples.

stability_fraction

Stability subsample fraction.

rank_by

Ranking method.

stability_weight

Stability penalty.

threshold_method

Threshold used for final class labels.

verbose

Logical.

Value

An object of class "automlr_binary_ensemble".

Fit a continuous-outcome prediction ensemble.

Description

Fit a continuous-outcome prediction ensemble.

Usage

fit_continuous_ensemble(
  X,
  y = NULL,
  params = continuousmlr_parameters(),
  algorithms = params$algorithms,
  min_r2 = params$min_r2_accept,
  auto_min_r2 = params$auto_min_r2 %||% FALSE,
  auto_quantile = params$auto_quantile %||% 0.5,
  strategy = c("best_subset", "threshold"),
  min_models = 1L,
  max_models = 2L,
  weight_method = c("inverse_rmse", "equal", "r2"),
  allow_same_algorithm = FALSE,
  max_failed_fraction = 0.2,
  rank_by = c("rmse", "r2"),
  verbose = params$verbose
)

Arguments

X

Feature matrix, or an "automlr_continuous_input" object.

y

Numeric outcome. Leave NULL when X is an input object.

params

Output of continuousmlr_parameters().

algorithms

Continuous algorithm keys.

min_r2

Minimum R-squared for threshold strategy.

auto_min_r2

Logical. If TRUE, compute min_r2 from the candidate R-squared distribution using auto_quantile.

auto_quantile

Quantile used for automatic threshold selection.

strategy

"best_subset" or "threshold".

min_models

Minimum combination size.

max_models

Maximum combination size.

weight_method

One of "inverse_rmse", "equal", or "r2".

allow_same_algorithm

Logical.

max_failed_fraction

Maximum failed sample fraction.

rank_by

Ranking method.

verbose

Logical.

Value

An object of class "automlr_continuous_ensemble".

Fit an ordinal-outcome ensemble.

Description

Fit an ordinal-outcome ensemble.

Usage

fit_ordinal_ensemble(
  X,
  y = NULL,
  params = ordinalmlr_parameters(),
  algorithms = params$algorithms,
  min_qwk = params$min_qwk_accept,
  auto_min_qwk = params$auto_min_qwk %||% FALSE,
  auto_quantile = params$auto_quantile %||% 0.5,
  strategy = c("best_subset", "threshold"),
  min_models = 1L,
  max_models = 2L,
  weight_method = c("qwk", "equal", "inverse_mae"),
  allow_same_algorithm = FALSE,
  rank_by = c("qwk", "class_mae"),
  verbose = params$verbose
)

Arguments

X

Feature matrix, or an "automlr_ordinal_input" object.

y

Ordinal integer outcome codes. Leave NULL when X is an input object.

params

Output of ordinalmlr_parameters().

algorithms

Ordinal algorithm keys.

min_qwk

Minimum quadratic weighted kappa for threshold strategy.

auto_min_qwk

Logical. If TRUE, compute min_qwk from the candidate kappa distribution using auto_quantile.

auto_quantile

Quantile used for automatic threshold selection.

strategy

"best_subset" or "threshold".

min_models

Minimum combination size.

max_models

Maximum combination size.

weight_method

One of "qwk", "equal", or "inverse_mae".

allow_same_algorithm

Logical.

rank_by

Ranking method.

verbose

Logical.

Value

An object of class "automlr_ordinal_ensemble".

Fit a weighted ensemble of survival-risk models.

Description

The ensemble first estimates each candidate model's LOOCV C-index. With the default "best_subset" strategy, it enumerates model subsets up to max_models variants and chooses the subset with the highest combined LOOCV C-index. With "threshold", it keeps all single models meeting min_cindex (or the best finite model if none meet it).

Usage

fit_surv_ensemble(
  X,
  y = NULL,
  params = automlr_parameters(),
  algorithms = params$algorithms,
  min_cindex = params$min_cindex_accept,
  auto_min_cindex = params$auto_min_cindex %||% FALSE,
  auto_quantile = params$auto_quantile %||% 0.5,
  strategy = c("best_subset", "threshold"),
  min_models = 1L,
  max_models = 2L,
  weight_method = c("cindex", "equal", "cindex_stability"),
  allow_same_algorithm = FALSE,
  max_failed_fraction = 0.2,
  stability_groups = NULL,
  stability_resamples = params$stability_resamples %||% 0L,
  stability_fraction = params$stability_fraction %||% 0.8,
  rank_by = c("cindex", "stability_weighted"),
  stability_weight = params$stability_weight %||% 0.1,
  diagnostic_times = NULL,
  verbose = params$verbose
)

Arguments

X

Feature matrix, or an "automlr_input" object returned by prepare_cohort_input().

y

survival::Surv object. Leave NULL when X is an "automlr_input" object.

params

Output of automlr_parameters().

algorithms

Character vector of registry keys.

min_cindex

Minimum C-index for automatic inclusion.

auto_min_cindex

Logical. If TRUE, compute min_cindex from the candidate LOOCV C-index distribution using auto_quantile.

auto_quantile

Quantile used for automatic threshold selection.

strategy

One of "best_subset" or "threshold".

min_models

Minimum subset size for "best_subset".

max_models

Maximum subset size for "best_subset".

weight_method

One of "cindex", "equal", or "cindex_stability".

allow_same_algorithm

Logical, allow multiple variants from the same base algorithm in one selected combination.

max_failed_fraction

Maximum allowed LOOCV fold failure fraction for candidates entering combination search.

stability_groups

Optional group/queue labels used for stability diagnostics.

stability_resamples

Number of repeated subsamples for stability diagnostics.

stability_fraction

Fraction of samples in each stability subsample.

rank_by

Ranking method passed to evaluate_surv_combinations().

stability_weight

Non-negative stability penalty multiplier used when rank_by = "stability_weighted".

diagnostic_times

Optional time points for time-dependent AUC diagnostics in the combination table.

verbose

Logical.

Value

A list of class "automlr_surv_ensemble".

Return the binary-classification algorithm registry.

Description

Return the binary-classification algorithm registry.

Usage

get_binary_registry()

Value

A named list of algorithm specs.

Return the continuous-outcome algorithm registry.

Description

Return the continuous-outcome algorithm registry.

Usage

get_continuous_registry()

Value

A named list of algorithm specs.

Return the ordinal-outcome algorithm registry.

Description

Return the ordinal-outcome algorithm registry.

Usage

get_ordinal_registry()

Value

A named list of algorithm specs.

Return the full survival-algorithm registry.

Description

Return the full survival-algorithm registry.

Usage

get_surv_registry()

Value

A named list of algorithm specs.

Enable file + console logging for the current R session.

Description

Writes a timestamped log file to log_dir (default: file.path(getwd(), "automlr_logs")) and keeps at most max_log_files.

Usage

initialize_auto_logging(log_dir = NULL, max_log_files = 10)

Arguments

log_dir

Directory to write logs into.

max_log_files

Retain only the N most recent logs.

Value

Invisibly returns the log4r logger when available, otherwise NULL.

List supported binary-classification algorithms.

Description

List supported binary-classification algorithms.

Usage

list_binary_algorithms()

Value

Character vector of algorithm keys.

List binary-classification model variants.

Description

List binary-classification model variants.

Usage

list_binary_model_variants(
  params = binarymlr_parameters(),
  algorithms = params$algorithms
)

Arguments

params

Output of binarymlr_parameters().

algorithms

Binary algorithm keys.

Value

A data.frame of concrete model variants.

List supported continuous-outcome algorithms.

Description

List supported continuous-outcome algorithms.

Usage

list_continuous_algorithms()

Value

Character vector of algorithm keys.

List continuous-outcome model variants.

Description

List continuous-outcome model variants.

Usage

list_continuous_model_variants(
  params = continuousmlr_parameters(),
  algorithms = params$algorithms
)

Arguments

params

Output of continuousmlr_parameters().

algorithms

Continuous algorithm keys.

Value

A data.frame of concrete model variants.

List concrete model variants generated from algorithm grids.

Description

Each row is one candidate model that can enter a model combination. Multiple rows can come from the same base algorithm when its registry grid contains multiple hyperparameter settings.

Usage

list_model_variants(
  params = automlr_parameters(),
  algorithms = params$algorithms
)

Arguments

params

Output of automlr_parameters().

algorithms

Character vector of registry keys.

Value

A data.frame of candidate model variants.

List supported ordinal-outcome algorithms.

Description

List supported ordinal-outcome algorithms.

Usage

list_ordinal_algorithms()

Value

Character vector of algorithm keys.

List ordinal-outcome model variants.

Description

List ordinal-outcome model variants.

Usage

list_ordinal_model_variants(
  params = ordinalmlr_parameters(),
  algorithms = params$algorithms
)

Arguments

params

Output of ordinalmlr_parameters().

algorithms

Ordinal algorithm keys.

Value

A data.frame of concrete model variants.

List the supported survival algorithms (keys).

Description

List the supported survival algorithms (keys).

Usage

list_surv_algorithms()

Value

Character vector of algorithm keys.

Leave-one-out cross-validation AUC for one binary algorithm.

Description

Leave-one-out cross-validation AUC for one binary algorithm.

Usage

loocv_auc(
  X,
  y,
  fit_fn,
  predict_fn,
  hparam = list(),
  seed = NULL,
  verbose = FALSE,
  n_cores = 1L
)

Arguments

X

Numeric feature matrix.

y

Binary 0/1 outcome.

fit_fn

Function ⁠(X_train, y_train, hparam) -> model⁠.

predict_fn

Function ⁠(model, X_test) -> positive-class probability⁠.

hparam

Hyperparameter list.

seed

Optional seed.

verbose

Logical.

n_cores

Integer number of fold workers.

Value

A list with AUC, PR-AUC, Brier score, probabilities, and failures.

Leave-one-out cross-validation C-index for one survival algorithm.

Description

Loops over every sample, fits the model on the remaining n-1, predicts the held-out one, then computes Harrell's C-index on the full vector of held-out risk scores. Any fold that errors records NA for that sample and the error message; the C-index is computed on the remaining folds.

Usage

loocv_cindex(
  X,
  y,
  fit_fn,
  predict_fn,
  hparam = list(),
  seed = NULL,
  verbose = FALSE,
  n_cores = 1L
)

Arguments

X

Numeric matrix or data.frame of features (⁠n x p⁠).

y

A survival::Surv object of length n.

fit_fn

Function ⁠(X_train, y_train, hparam) -> model⁠.

predict_fn

Function ⁠(model, X_test) -> numeric risk score⁠ (one per row of X_test). Higher score = higher hazard.

hparam

Named list of hyperparameters passed to fit_fn. Use list() for algorithms that tune internally.

seed

Integer seed set once before the loop for reproducibility; NULL leaves the RNG untouched.

verbose

Logical, print a progress message every ~10%.

n_cores

Integer, number of workers for fold-level parallelism.

Value

A list with components:

cindex

Scalar Harrell's C-index on the aggregated predictions.

risk

Numeric vector of length n with per-sample LOOCV risk scores (NA where the fold errored).

n_folds

n.

n_failed

Number of folds that errored.

errors

Character vector of error messages (may be length 0).

elapsed_sec

Wall-clock seconds.

Accuracy for ordinal class predictions.

Description

Accuracy for ordinal class predictions.

Usage

ordinal_accuracy(y, pred_class)

Arguments

y

Observed positive integer level codes.

pred_class

Predicted positive integer level codes.

Value

A numeric scalar.

Balanced accuracy for ordinal class predictions.

Description

Balanced accuracy for ordinal class predictions.

Usage

ordinal_balanced_accuracy(y, pred_class)

Arguments

y

Observed positive integer level codes.

pred_class

Predicted positive integer level codes.

Value

A numeric scalar.

Mean absolute class error for ordinal predictions.

Description

Mean absolute class error for ordinal predictions.

Usage

ordinal_mae(y, pred_class)

Arguments

y

Observed positive integer level codes.

pred_class

Predicted positive integer level codes.

Value

A numeric scalar.

Quadratic weighted kappa for ordinal predictions.

Description

Quadratic weighted kappa for ordinal predictions.

Usage

ordinal_qwk(y, pred_class, n_levels = max(y, pred_class, na.rm = TRUE))

Arguments

y

Observed positive integer level codes.

pred_class

Predicted positive integer level codes.

n_levels

Number of ordered outcome levels.

Value

A numeric scalar.

Default parameters for AutoMLR ordinal-outcome workflows.

Description

Default parameters for AutoMLR ordinal-outcome workflows.

Usage

ordinalmlr_parameters(
  seed = 123L,
  algorithms = NULL,
  resampling = "loocv",
  k_folds = 5L,
  repeats = 1L,
  min_qwk_accept = 0,
  auto_min_qwk = FALSE,
  auto_quantile = 0.5,
  missing_fraction_cutoff = 0.2,
  screen_by_variance = TRUE,
  variance_quantile_cutoff = 0,
  standardize_features = FALSE,
  n_cores = 1L,
  verbose = TRUE
)

Arguments

seed

Base random seed.

algorithms

Character vector of ordinal algorithm keys.

resampling

Resampling scheme.

k_folds

Number of folds for k-fold CV.

repeats

Number of repeats for repeated k-fold CV.

min_qwk_accept

Minimum quadratic weighted kappa for threshold strategy.

auto_min_qwk

Logical. If TRUE and fit_ordinal_ensemble() uses strategy = "threshold", replace min_qwk_accept with a recommended cutoff computed from candidate-model kappa values.

auto_quantile

Quantile used when an automatic threshold is requested. 0.50 uses the median candidate metric; larger values are more selective.

missing_fraction_cutoff

Drop features above this missing fraction.

screen_by_variance

Logical, drop zero / low-variance features.

variance_quantile_cutoff

Optional lower variance quantile to drop.

standardize_features

Logical, center and scale features.

n_cores

Integer, number of fold workers.

verbose

Logical, print progress.

Value

A named list.

Parallel lapply that transparently falls back to sequential.

Description

Parallel lapply that transparently falls back to sequential.

Usage

parallel_lapply(X, FUN, ..., cores = NULL)

Arguments

X

Input list / vector.

FUN

Function.

...

Extra args to FUN.

cores

Optionally override the global core count.

Value

List of results.

Predict binary ensemble probabilities or classes.

Description

Predict binary ensemble probabilities or classes.

Usage

## S3 method for class 'automlr_binary_ensemble'
predict(
  object,
  newX,
  type = c("prob", "class"),
  threshold = object$threshold,
  ...
)

Arguments

object

Object returned by fit_binary_ensemble().

newX

Feature matrix.

type

"prob" for positive-class probability or "class" for 0/1.

threshold

Optional class threshold.

...

Ignored.

Value

Numeric vector.

Predict continuous ensemble values.

Description

Predict continuous ensemble values.

Usage

## S3 method for class 'automlr_continuous_ensemble'
predict(object, newX, ...)

Arguments

object

Object returned by fit_continuous_ensemble().

newX

Feature matrix.

...

Ignored.

Value

Numeric vector.

Predict ordinal ensemble scores or classes.

Description

Predict ordinal ensemble scores or classes.

Usage

## S3 method for class 'automlr_ordinal_ensemble'
predict(object, newX, type = c("score", "code", "class"), ...)

Arguments

object

Object returned by fit_ordinal_ensemble().

newX

Feature matrix.

type

"score", "code", or "class".

...

Ignored.

Value

Numeric scores/codes or class labels.

Predict weighted ensemble risk.

Description

Predict weighted ensemble risk.

Usage

## S3 method for class 'automlr_surv_ensemble'
predict(object, newX, ...)

Arguments

object

An object returned by fit_surv_ensemble().

newX

Feature matrix.

...

Ignored.

Value

Numeric risk score; higher means higher predicted hazard.

Prepare multi-cohort binary-classification data.

Description

Splits a long-format data frame by cohort, maps the outcome to 0/1, keeps numeric shared features across cohorts, and returns an object for binary AutoMLR workflows.

Usage

prepare_binary_cohort_input(
  data,
  cohort,
  outcome,
  id = NULL,
  positive_class = 1,
  negative_class = NULL,
  collapse_other = FALSE,
  drop_cohorts = NULL
)

Arguments

data

A data.frame with one row per sample.

cohort

Name of the cohort / dataset column.

outcome

Name of the binary outcome column.

id

Optional sample-id column.

positive_class

Value in outcome that should be treated as the positive class. Defaults to 1.

negative_class

Optional value in outcome that should be treated as the negative class. When NULL, the remaining level is inferred only if the outcome has exactly two non-missing levels.

collapse_other

Logical. If TRUE, all classes other than positive_class are treated as negative. Defaults to FALSE to avoid accidental multi-class collapse.

drop_cohorts

Optional cohorts to exclude.

Value

An object of class "automlr_binary_input".

Prepare multi-cohort survival data from a single long-format table.

Description

Splits data by the cohort column and computes the feature intersection across cohorts. Returns a tidy object for downstream fitting plus diagnostic info for the user.

Usage

prepare_cohort_input(
  data,
  cohort,
  time,
  status,
  id = NULL,
  drop_cohorts = NULL
)

Arguments

data

A data.frame with one row per sample.

cohort

Name of the column identifying cohort membership.

time

Name of the survival time column (numeric, > 0).

status

Name of the event indicator column (0/1; 1 = event).

id

Optional name of a sample-id column; just passed through.

drop_cohorts

Optional character vector of cohorts to exclude.

Value

An S3 list of class "automlr_input" with components:

cohorts

Named list of per-cohort data frames restricted to shared_features + time + status (+ id).

shared_features

Character vector of feature columns present in every cohort (the intersection).

per_cohort_features

Named list of each cohort's raw feature set.

dropped_features

Features present in at least one cohort but not all, therefore excluded from the intersection.

summary

data.frame: cohort, n_samples, n_events, median_time, n_raw_features, n_shared_features.

meta

Echo of column names used.

Prepare multi-cohort continuous-outcome data.

Description

Splits a long-format data frame by cohort, keeps numeric shared features, and returns an object for continuous AutoMLR workflows.

Usage

prepare_continuous_cohort_input(
  data,
  cohort,
  outcome,
  id = NULL,
  drop_cohorts = NULL
)

Arguments

data

A data.frame with one row per sample.

cohort

Name of the cohort / dataset column.

outcome

Name of the numeric outcome column.

id

Optional sample-id column.

drop_cohorts

Optional cohorts to exclude.

Value

An object of class "automlr_continuous_input".

Prepare multi-cohort ordinal-outcome data.

Description

Maps an ordered outcome to integer scores, keeps numeric shared features, and returns an object for ordinal AutoMLR workflows.

Usage

prepare_ordinal_cohort_input(
  data,
  cohort,
  outcome,
  ordered_levels = NULL,
  id = NULL,
  drop_cohorts = NULL
)

Arguments

data

A data.frame with one row per sample.

cohort

Name of the cohort / dataset column.

outcome

Name of the ordinal outcome column.

ordered_levels

Optional ordered outcome levels from low to high.

id

Optional sample-id column.

drop_cohorts

Optional cohorts to exclude.

Value

An object of class "automlr_ordinal_input".

Print an AutoMLR dependency report.

Description

Print an AutoMLR dependency report.

Usage

## S3 method for class 'automlr_dependency_report'
print(x, ...)

Arguments

x

An object returned by check_automlr_dependencies().

...

Unused.

Value

Invisibly returns x.

Print method for extreme survival screening

Description

Print method for extreme survival screening

Usage

## S3 method for class 'automlr_extreme_screen'
print(x, ...)

Arguments

x

An object returned by extreme_surv_screen().

...

Ignored.

Value

Invisibly returns x.

Recommend a binary AUC cutoff from candidate model results.

Description

Uses the finite receiver operating characteristic area under the curve (AUC) values in a binary candidate summary and returns the requested quantile as a threshold for "threshold" strategy model selection.

Usage

recommend_binary_auc_threshold(loocv_set, auto_quantile = 0.5, minimum = 0.5)

Arguments

loocv_set

An object returned by evaluate_binary_algorithms_loocv(), or a data frame containing a numeric auc column.

auto_quantile

Numeric quantile in ⁠[0, 1]⁠. Larger values are more selective; 0.50 uses the median candidate AUC.

minimum

Lower bound for the returned threshold. Defaults to 0.5.

Value

A numeric scalar giving the recommended minimum AUC.

Recommend a continuous R-squared cutoff from candidate model results.

Description

Uses finite out-of-fold R-squared values and returns the requested quantile as a threshold for "threshold" strategy model selection.

Usage

recommend_continuous_r2_threshold(
  resample_set,
  auto_quantile = 0.5,
  minimum = 0
)

Arguments

resample_set

An object returned by evaluate_continuous_algorithms(), or a data frame containing a numeric r2 column.

auto_quantile

Numeric quantile in ⁠[0, 1]⁠. Larger values are more selective; 0.50 uses the median candidate R-squared.

minimum

Lower bound for the returned threshold. Defaults to 0.

Value

A numeric scalar giving the recommended minimum R-squared.

Recommend an ordinal kappa cutoff from candidate model results.

Description

Uses finite out-of-fold quadratic weighted kappa values and returns the requested quantile as a threshold for "threshold" strategy model selection.

Usage

recommend_ordinal_qwk_threshold(resample_set, auto_quantile = 0.5, minimum = 0)

Arguments

resample_set

An object returned by evaluate_ordinal_algorithms(), or a data frame containing a numeric qwk column.

auto_quantile

Numeric quantile in ⁠[0, 1]⁠. Larger values are more selective; 0.50 uses the median candidate kappa.

minimum

Lower bound for the returned threshold. Defaults to 0.

Value

A numeric scalar giving the recommended minimum quadratic weighted kappa.

Recommend a survival C-index cutoff from candidate model results.

Description

Uses the finite leave-one-out cross-validation concordance index (C-index) values in a survival candidate summary and returns the requested quantile as a threshold for "threshold" strategy model selection.

Usage

recommend_surv_cindex_threshold(loocv_set, auto_quantile = 0.5, minimum = 0.5)

Arguments

loocv_set

An object returned by evaluate_algorithms_loocv(), or a data frame containing a numeric cindex column.

auto_quantile

Numeric quantile in ⁠[0, 1]⁠. Larger values are more selective; 0.50 uses the median candidate C-index.

minimum

Lower bound for the returned threshold. Defaults to 0.5 so models below random concordance are not automatically accepted.

Value

A numeric scalar giving the recommended minimum C-index.

Render an HTML report for a fitted binary ensemble.

Description

Render an HTML report for a fitted binary ensemble.

Usage

render_binary_report(
  object,
  output_dir = "automlr_binary_report",
  report_file = "index.html",
  title = "AutoMLR Binary Report",
  top_n = 20L,
  overwrite = TRUE,
  summary_language = c("bilingual", "en", "zh")
)

Arguments

object

Object returned by fit_binary_ensemble().

output_dir

Report directory.

report_file

HTML file name.

title

Report title.

top_n

Number of top rows.

overwrite

Logical.

summary_language

"bilingual", "en", or "zh".

Value

Invisibly returns report path.

Render an HTML report for a fitted continuous ensemble.

Description

Render an HTML report for a fitted continuous ensemble.

Usage

render_continuous_report(
  object,
  output_dir = "automlr_continuous_report",
  report_file = "index.html",
  title = "AutoMLR Continuous Report",
  top_n = 20L,
  overwrite = TRUE
)

Arguments

object

Object returned by fit_continuous_ensemble().

output_dir

Report directory.

report_file

HTML file name.

title

Report title.

top_n

Number of top rows.

overwrite

Logical.

Value

Invisibly returns report path.

Render an HTML report for a fitted ordinal ensemble.

Description

Render an HTML report for a fitted ordinal ensemble.

Usage

render_ordinal_report(
  object,
  output_dir = "automlr_ordinal_report",
  report_file = "index.html",
  title = "AutoMLR Ordinal Report",
  top_n = 20L,
  overwrite = TRUE
)

Arguments

object

Object returned by fit_ordinal_ensemble().

output_dir

Report directory.

report_file

HTML file name.

title

Report title.

top_n

Number of top rows.

overwrite

Logical.

Value

Invisibly returns report path.

Render an HTML report for a fitted survival ensemble.

Description

Writes a self-contained HTML summary plus separate ⁠figures/⁠ and ⁠tables/⁠ folders. Model selection remains whatever was used by fit_surv_ensemble(); this function only reports diagnostics.

Usage

render_surv_report(
  object,
  output_dir = "automlr_report",
  report_file = "index.html",
  title = "AutoMLR Survival Report",
  top_n = 20L,
  overwrite = TRUE,
  summary_language = c("bilingual", "en", "zh")
)

Arguments

object

An object returned by fit_surv_ensemble().

output_dir

Directory where the report folder should be written.

report_file

HTML file name.

title

Report title.

top_n

Number of top single models / combinations to show.

overwrite

Logical, overwrite existing report files.

summary_language

Language used in summary_report.md, either "bilingual", "en", or "zh".

Value

Invisibly returns the HTML report path.

Print a binary cohort-intersection report.

Description

Print a binary cohort-intersection report.

Usage

report_binary_cohort_intersection(input)

Arguments

input

An object returned by prepare_binary_cohort_input().

Value

Invisibly returns input.

Print a human-readable report of the cohort intersection.

Description

Tells the user:

Usage

report_cohort_intersection(input)

Arguments

input

An object returned by prepare_cohort_input().

Value

Invisibly returns input.

Print a continuous cohort-intersection report.

Description

Print a continuous cohort-intersection report.

Usage

report_continuous_cohort_intersection(input)

Arguments

input

An object returned by prepare_continuous_cohort_input().

Value

Invisibly returns input.

Print an ordinal cohort-intersection report.

Description

Print an ordinal cohort-intersection report.

Usage

report_ordinal_cohort_intersection(input)

Arguments

input

An object returned by prepare_ordinal_cohort_input().

Value

Invisibly returns input.

Start the parallel backend.

Description

Thin wrapper around future::plan() — see ?future::plan. If the optional future package is unavailable, AutoMLR keeps using sequential execution.

Usage

start_parallel(
  cores = get_parallel_cores(),
  strategy = c("multisession", "multicore")
)

Arguments

cores

Integer, number of workers.

strategy

One of "multisession" (default, cross-platform) or "multicore" (Linux/macOS only).

Value

Invisibly TRUE when the backend was started, or FALSE when the optional future package is unavailable.

Stop the parallel backend.

Description

Stop the parallel backend.

Usage

stop_parallel()

Value

Invisibly TRUE when a future backend was reset, or FALSE when the optional future package is unavailable.

Summarize base-model screening results in Markdown

Description

Summarize base-model screening results in Markdown

Usage

summarize_base_models(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

An object returned by fit_surv_ensemble().

top_n

Number of top model rows to show.

language

Summary language: "bilingual", "en", or "zh".

Value

A Markdown string.

Summarize a complete binary AutoMLR analysis in Markdown.

Description

Summarize a complete binary AutoMLR analysis in Markdown.

Usage

summarize_binary_analysis_results(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

A fitted binary ensemble.

top_n

Number of rows per section.

language

"bilingual", "en", or "zh".

Value

Markdown string.

Summarize binary base-model screening in Markdown.

Description

Summarize binary base-model screening in Markdown.

Usage

summarize_binary_base_models(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

A fitted binary ensemble.

top_n

Number of rows.

language

"bilingual", "en", or "zh".

Value

Markdown string.

Summarize binary data preparation in Markdown.

Description

Summarize binary data preparation in Markdown.

Usage

summarize_binary_data_preparation(
  x,
  top_n = 10L,
  language = c("bilingual", "en", "zh")
)

Arguments

x

An automlr_binary_input or fitted binary ensemble.

top_n

Number of rows.

language

"bilingual", "en", or "zh".

Value

Markdown string.

Summarize binary ensemble selection in Markdown.

Description

Summarize binary ensemble selection in Markdown.

Usage

summarize_binary_ensemble_results(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

A fitted binary ensemble.

top_n

Number of rows.

language

"bilingual", "en", or "zh".

Value

Markdown string.

Summarize binary explainability outputs in Markdown.

Description

Summarize binary explainability outputs in Markdown.

Usage

summarize_binary_explainability_results(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

A fitted binary ensemble.

top_n

Number of rows.

language

"bilingual", "en", or "zh".

Value

Markdown string.

Summarize data-preparation results in Markdown

Description

Creates a bilingual or single-language summary of cohort/sample/event counts, shared features, dropped non-shared features, and simple data-risk notes.

Usage

summarize_data_preparation(
  x,
  top_n = 10L,
  language = c("bilingual", "en", "zh")
)

Arguments

x

An automlr_input object or an object returned by fit_surv_ensemble().

top_n

Number of cohort rows to show.

language

Summary language: "bilingual", "en", or "zh".

Value

A Markdown string.

Summarize ensemble-selection results in Markdown

Description

Summarize ensemble-selection results in Markdown

Usage

summarize_ensemble_results(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

An object returned by fit_surv_ensemble().

top_n

Number of top combination rows to show.

language

Summary language: "bilingual", "en", or "zh".

Value

A Markdown string.

Summarize explainability and clinical-utility outputs in Markdown

Description

Summarize explainability and clinical-utility outputs in Markdown

Usage

summarize_explainability_results(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

An object returned by fit_surv_ensemble().

top_n

Number of top feature/model rows to show.

language

Summary language: "bilingual", "en", or "zh".

Value

A Markdown string.

Summarize extreme-screening results in readable Markdown

Description

Creates a compact interpretation of an extreme_surv_screen() result: apparent-screen leaders, best seed-search model, top seed rows, best rows after seed de-duplication, combination-level stability, and failure notes.

Usage

summarize_extreme_screen_results(
  x,
  top_n = 3L,
  language = c("bilingual", "en", "zh")
)

Arguments

x

An object returned by extreme_surv_screen().

top_n

Number of rows to include in each top-results section.

language

Summary language, either "bilingual", "zh", or "en".

Value

A single Markdown string.

Summarize a complete regular survival AutoML analysis in Markdown

Description

Combines data-preparation, base-model, and ensemble-selection summaries. This is the regular-analysis counterpart to summarize_extreme_screen_results().

Usage

summarize_surv_analysis_results(
  object,
  top_n = 5L,
  language = c("bilingual", "en", "zh")
)

Arguments

object

An object returned by fit_surv_ensemble().

top_n

Number of top rows to show per section.

language

Summary language: "bilingual", "en", or "zh".

Value

A Markdown string.