Multiple Randomised Cross Validation: private API

abstract type SubsetSizeEstimator <: AbstractEstimator

Abstract supertype for estimators that determine the size of each asset subset.

A subtype reaches a MultipleRandomised through its subset_size field, and is called on the returns data to give the value.

Related

• MultipleRandomised
• SubsetSizeEC

abstract type NumberSubsetsEstimator <: AbstractEstimator

Abstract supertype for estimators that determine the number of random subsets to draw.

A subtype reaches a MultipleRandomised through its n_subsets field, and is called on the returns data to give the value.

Related

• MultipleRandomised
• NumberSubsetsEC

abstract type WindowSizeEstimator <: AbstractEstimator

Abstract supertype for estimators that determine the rolling window size.

A subtype reaches a MultipleRandomised through its window_size field, and is called on the returns data to give the value.

Related

• MultipleRandomised
• WindowSizeEC

const SubsetSizeEC

Union of SubsetSizeEstimator subtypes and plain functions that compute a subset size from a returns dataset.

const NumberSubsetsEC

Union of NumberSubsetsEstimator subtypes and plain functions that compute the number of subsets from a returns dataset.

const WindowSizeEC

Union of WindowSizeEstimator subtypes and plain functions that compute a window size from a returns dataset.

const SubsetSizeE

Union of a concrete subset-size value or an estimator/function for it.

const NumberSubsetsE

Union of a concrete number-of-subsets value or an estimator/function for it.

const WindowSizeE

Union of a concrete window-size value or an estimator/function for it.

const MRCVR = Union{<:MultipleRandomised, <:MultipleRandomisedResult}

Alias for a multiple-randomised cross-validation estimator or result.

Matches either a MultipleRandomised estimator or a MultipleRandomisedResult.

Related

• MultipleRandomised
• MultipleRandomisedResult

path_fit_and_predict(opt, rd, train_idx, test_idx, cols; ex, id, cv)

Fit and predict along a sequence of (train, test, asset) triples, respecting sequential constraints.

The path runs through fold_loop, which takes each fold's asset-subset view of (opt, rd) and resolves the fold's time-dependent entries. The path runs sequentially when the optimiser needs the previous fold's weights, and in parallel over ex otherwise. A time-dependent optimiser alone does not force sequential execution, because its per-fold values are known upfront. cv is the scheme the path belongs to, handed on so the loop reads its Fold Fit (fold_fit): under an OnlineStep the path's estimator is sliced to the subset once and threaded through the path's folds.

Arguments

• opt::NonFiniteAllocationOptimisationEstimator: Portfolio optimisation estimator.
• rd::ReturnsResult: FullMoment returns data.
• train_idx: Sequence of training index vectors.
• test_idx: Sequence of test index vectors.
• cols: Sequence of asset column indices for each fold.
• ex::FLoops.Transducers.Executor: Executor for parallel processing.
• id: Optional path identifier.
• cv: The scheme the path belongs to, or nothing.

Returns

• MultiPeriodPredictionResult with predictions sorted by test index.

Related

• fit_and_predict
• fold_loop
• needs_previous_weights
• is_time_dependent
• update_time_dependent_estimator

combination_by_index(idx::Integer, N::Integer, k::Integer) -> VecInt

Return the idx-th combination of k items from N total items.

Internal helper for combinatorial path generation. Converts a lexicographic combination index to the actual combination elements.

Arguments

• idx: Combination index (1-based).
• N: Total number of items.
• k: Number of items in each combination.

Returns

• Vector of k item indices.

Related

• CombinatorialCrossValidation
• sample_unique_assets

get_window_size(window_size, rd, args...)

Get the actual rolling window size for multiple-randomised cross-validation.

Resolves the window size from nothing (no windowing), an integer (direct count), a float (fraction of observations), or a callable.

Arguments

• window_size: Window size specification (nothing, integer, float, or callable).
• rd: Returns result or prior.
• args...: Additional arguments.

Returns

• Integer window size or nothing.

Related

• MultipleRandomised
• get_subset_size

get_n_subsets(n_subsets, args...)

Get the number of asset subsets for multiple-randomised cross-validation.

Resolves the number of subsets from either an integer (direct count) or a callable that computes it from the returns data.

This is the single point at which the (possibly TimeDependent or callable) subset count becomes a concrete integer, so it is where the RESOURCE_LIMITS max_n_subsets ceiling is enforced — a constructor check could not see the value a schedule or callable ultimately produces. Every subset runs a full inner optimisation, so an absurd count is a compute-exhaustion sink rather than a merely large allocation.

Arguments

• n_subsets: Integer or callable number-of-subsets specification.
• args...: Additional arguments (returns result or prior).

Returns

• Integer number of subsets.

Related

• MultipleRandomised
• sample_unique_assets
• assert_resource_cap

sample_unique_assets(N, k, n_subsets; kwargs...)

Sample n_subsets unique asset subsets of size k from N assets.

Internal function used in multiple-randomised cross-validation to generate diverse asset subsets for resampling.

Arguments

• N: Total number of assets.
• k: Subset size.
• n_subsets: Number of unique subsets to sample.
• kwargs...: Additional keyword arguments (e.g., random seed).

Returns

• Matrix of size (k, n_subsets) with asset indices.

Related

• MultipleRandomised
• combination_by_index

get_subset_size(subset_size, rd, args...)

Get the actual asset subset size for multiple-randomised cross-validation.

Resolves the subset size from either an integer (direct count) or a fraction of the total assets.

Arguments

• subset_size: Integer or float subset size specification.
• rd: Returns result or prior.
• args...: Additional arguments.

Returns

• Integer subset size.

Related

• MultipleRandomised
• sample_unique_assets