xval_SVR

Object methods index · Toolbox folders

End-to-end SVR pipeline for continuous outcomes, with the same scaffolding as xval_SVM: stratified holdouts that respect repeated-measures groupings, optional nested hyperparameter optimisation, repeated random splits, and bootstrap inference on feature weights. Use this when the target is continuous (e.g. pain rating, age, behavioural score) and you want a defensible, fully cross-validated linear regression.

Code map

Editable PowerPoint version

Usage

S = xval_SVR(X, Y, id, varargin)

Inputs

Argument	Type	Description
X	[N × p] numeric	Predictor matrix (observations × features).
Y	[N × 1] numeric	Continuous outcome to predict.
id	[N × 1] numeric	Grouping codes (e.g. subject id). All observations sharing an id are kept together in train or test. Use 1:N or [] for no grouping.
'doplot', logical	flag	Create plots. Default true. Use 'noplot' to suppress.
'doverbose', logical	flag	Verbose output. Default true. Use 'noverbose' to suppress.
'dooptimize', logical	flag	Nested hyperparameter optimisation via Bayesian search. Default true. Use 'nooptimize' to skip.
'doprepeats', integer	param	Number of repeated cross-validations with different partitions. Default 10. Use 'norepeats' to skip.
'dobootstrap' / 'nobootstrap' / 'nboot', integer	flag / param	Bootstrap feature weights for inference. Default on. 'nboot', N sets number of bootstrap samples.
'modeloptions', cell	param	Cell array of name/value pairs forwarded to fitrsvm. Default {'KernelFunction', 'linear'}.

Outputs

S is a structure with (among others):

Field	Description
Y, yfit	True and cross-validated predicted continuous outcomes.
id	Grouping variable.
w, b	Final-model weights and bias.
SVRModel	The full-data RegressionSVM object (Beta, Bias, KernelParameters.Scale).
nfolds, cvpartition, trIdx, teIdx	CV bookkeeping.
dist_from_hyperplane_xval	Cross-validated continuous score.
class_probability_xval	Platt-scaled probability (legacy field; see Notes).
crossval_accuracy	Cross-validated prediction r² (no hyperparam opt).
prediction_outcome_r	Simple prediction-outcome correlation (not for use as quantitative objective).
regression_d_singleinterval	Cohen's d effect size derived from r.
crossval_accuracy_opt_hyperparams	Accuracy with optimised hyperparameters (when dooptimize).
Y_within_id, scores_within_id, scorediff	Within-person reorganisation for paired tests.
crossval_accuracy_within, classification_d_within	Within-person metrics.
boot_w_mean, boot_w_ste, wZ, wP, wP_fdr_thr, boot_w_fdrsig, w_thresh_fdr	Bootstrap inference on feature weights, including FDR-corrected significance.
accfun	Function handle for accuracy computation.

Notes

• Linear fitrsvm only by default; the source has commented hooks for nonlinear kernels.
• Hyperparameter optimisation uses Bayesian search with a 5-fold inner CV (not grouped by id) and the smooth best-estimate criterion. Needs reasonably large samples to be useful.
• If you optimise hyperparameters AND repeat cross-validation, you get nested cross-validation — accurate but potentially slow.
• The class_probability_xval field is a holdover from the SVM scaffolding; for regression, the most useful continuous score is S.yfit (or S.dist_from_hyperplane_xval, which is the same up to a constant in the regression case).
• prediction_outcome_r is provided for reporting but should not be used as the optimisation objective — it is overly forgiving when predictions have the right rank but wrong scale. Use crossval_accuracy (prediction r²) for that. See Scheinost et al. 2019 for the rationale.

Example

% One observation per person, true linear signal + heavy noise
n = 50;                                       % participants
k = 120;                                      % features
Y = randn(n, 1);                              % True continuous outcome
X = Y * randn(1, k) + ones(n, 1) * randn(1, k) + 5 * randn(n, k);
id = (1:n)';                                  % One observation per id

% Quick cross-validated performance
S = xval_SVR(X, Y, id, 'nooptimize', 'norepeats', 'nobootstrap');

% Two observations per participant
id2 = [(1:n/2)'; (1:n/2)'];
S2 = xval_SVR(X, Y, id2, 'nooptimize', 'norepeats', 'nobootstrap');

Other examples

% Quick bootstrap test (small N — not for final inference)
S = xval_SVR(X, Y, id, 'nooptimize', 'norepeats', 'nboot', 100);

% Repeated cross-val only, 5 repeats
S = xval_SVR(X, Y, id, 'nooptimize', 'dorepeats', 5, 'nobootstrap');

% Nested hyperparameter optimisation only
S = xval_SVR(X, Y, id, 'norepeats', 'nobootstrap');

% Full pipeline: optimise, repeat, bootstrap
S = xval_SVR(X, Y, id);

% Silent run for batch use
S = xval_SVR(X, Y, id, 'nooptimize', 'norepeats', 'nobootstrap', ...
    'noverbose', 'noplot');

See also

• xval_SVM — binary SVM with the same scaffolding
• xval_classify — multi-class linear discriminant
• xval_select_holdout_set — covariate-balanced holdout sets
• fmri_data.predict — top-level CV prediction on imaging objects
• fmri_data.regress — voxelwise multiple regression (mass-univariate alternative)