PortfolioTesteR (version 0.1.4)

cv_tune_seq: Purged/embargoed K-fold CV for sequence models (inside IS window)

Description

Purged/embargoed K-fold CV for sequence models (inside IS window)

Usage

cv_tune_seq(
  features_list,
  labels,
  is_start,
  is_end,
  steps_grid = c(12L, 26L, 52L),
  horizon = 4L,
  fit_fn,
  predict_fn,
  k = 3L,
  purge = NULL,
  embargo = NULL,
  group = c("pooled"),
  max_train_samples = NULL,
  max_val_samples = NULL,
  na_action = c("omit", "zero"),
  metric = c("spearman", "rmse")
)

Value

data.table with columns like steps, folds, and CV score.

Arguments

features_list

List of feature panels (Date + symbols) for sequences.

labels

Label panel aligned to features (Date + symbols).

is_start, is_end

Inclusive IS window indices (on aligned dates).

steps_grid

Integer vector of candidate sequence lengths.

horizon

Forecast horizon (periods ahead).

fit_fn

Function (X, y) -> model for sequences.

predict_fn

Function (model, Xnew) -> numeric scores.

k

Number of CV folds.

purge

Gap (obs) removed between train/val within folds (default uses steps).

embargo

Embargo (obs) after validation to avoid bleed (default uses horizon).

group

'pooled', 'per_symbol', or 'per_group'.

max_train_samples

Optional cap on IS samples per fold.

max_val_samples

Optional cap on validation samples per fold.

na_action

How to handle NA features ('omit' or 'zero').

metric

IC metric ('spearman' or 'pearson').