Skip to contents

K-Fold Cross-Validation for Time Series

k_fold_cv_ts.Rd

Time-series-aware k-fold cross-validation over a grid of candidate p values, using a user-specified model and scoring function.

Usage

k_fold_cv_ts(
  x,
  y,
  k = 5,
  p_grid = seq(0.01, 0.5, by = 0.01),
  model_func = default_model_func,
  score_func = mse
)

Arguments

x

Numeric matrix or vector of time series data (rows = time points, columns = variables).

y

Numeric vector of target values to predict (must match number of rows of x).

k

Integer number of folds.

p_grid

Numeric vector of candidate p values to evaluate.

model_func

A function that accepts training x_train, y_train and returns a fitted model object.

score_func

A scoring function of the form function(preds, y_val) returning a numeric score.

Value

Optimal p value that minimizes the average validation score.

Details

This function partitions the time series into contiguous validation folds, fits the model on the training set for each fold, and evaluates the predictive performance on the validation set using score_func.

Examples

# Generate toy time-series data
set.seed(123)
x <- matrix(rnorm(200), ncol = 2)
y <- 0.5 * x[,1] - 0.3 * x[,2] + rnorm(100)

# Perform time-series k-fold CV using default model and scoring functions
best_p <- k_fold_cv_ts(
  x = x,
  y = y,
  k = 5,
  p_grid = seq(0.01, 0.5, by = 0.01),
  model_func = default_model_func,
  score_func = mse
)
print(best_p)
#> [1] 0.01