patterncausality (version 0.2.1)

pcCrossValidation: Pattern Causality Cross-Validation Analysis

Description

Evaluates the robustness of pattern causality measures through repeated sampling analysis. This function performs cross-validation by analyzing multiple subsets of the data to assess the stability of causality relationships.

Usage

pcCrossValidation(
  X,
  Y,
  E,
  tau,
  metric = "euclidean",
  h,
  weighted,
  distance_fn = NULL,
  state_space_fn = NULL,
  numberset,
  random = TRUE,
  bootstrap = 1,
  verbose = FALSE,
  n_cores = 1,
  relative = TRUE
)

Value

A pc_cv object containing:

• samples: Vector of sample sizes used

• results: Array of causality results

• parameters: List of analysis parameters

The results array structure depends on the bootstrap parameter:

• If bootstrap>1: A three-dimensional array where first dimension represents sample sizes, second dimension contains statistics (mean, quantiles, median), and third dimension represents causality types (positive, negative, dark)

• If bootstrap=1: A three-dimensional array where first dimension represents sample sizes, second dimension contains single values, and third dimension represents causality types (positive, negative, dark)

Arguments

X

Numeric vector representing the first time series.

Y

Numeric vector representing the second time series.

E

Integer specifying the embedding dimension.

tau

Integer specifying the time delay.

metric

Character string specifying the distance metric to use.

h

Integer specifying the prediction horizon.

weighted

Logical indicating whether to use weighted calculations.

distance_fn

Optional custom distance function.

state_space_fn

Optional custom state space function.

numberset

Numeric vector of sample sizes to analyze.

random

Logical indicating whether to use random sampling (default: TRUE).

bootstrap

Integer specifying the number of bootstrap iterations (default: 1).

verbose

Logical indicating whether to display progress messages.

n_cores

Integer specifying the number of cores to use for parallel computation (default: 1).

relative

Logical; if TRUE calculates relative changes ((new-old)/old), if FALSE calculates absolute changes (new-old) in signature space. Default is TRUE.

Details

Perform Pattern Causality Cross-Validation Analysis

The function implements these key steps:

• Validates input parameters and data

• Performs stratified sampling of time series data

• When random=TRUE and bootstrap>1, performs bootstrap sampling

• Computes pattern causality measures for each sample

• Aggregates results across all samples

When bootstrap sampling is enabled (random=TRUE and bootstrap>1), the function returns statistics including mean, 5% quantile, 95% quantile, and median for each sample size.

See Also

plot.pc_cv for visualizing cross-validation results print.pc_cv for printing cross-validation results summary.pc_cv for summarizing cross-validation results

Examples

# \donttest{
data(climate_indices)
X <- climate_indices$AO
Y <- climate_indices$AAO

# Basic cross-validation
cv_result <- pcCrossValidation(
  X, Y, 
  E = 3, tau = 1,
  metric = "euclidean",
  h = 1,
  weighted = FALSE,
  numberset = c(100, 200, 300)
)

# Cross-validation with bootstrap
cv_result_boot <- pcCrossValidation(
  X, Y,
  E = 3, tau = 1,
  metric = "euclidean",
  h = 1,
  weighted = FALSE,
  numberset = c(100, 200, 300),
  random = TRUE,
  bootstrap = 100
)
# }