05. Extending DataSplits

Custom Clustering

To add a new clustering algorithm:

1. Subtype ClusteringResult.
2. Implement assignments, nclusters, counts, wcounts for your result type.
3. Register a clustering function returning your result.

Example:

struct MyClusteringResult <: ClusteringResult
  assignments::Vector{Int}
end
assignments(r::MyClusteringResult) = r.assignments
nclusters(r::MyClusteringResult) = maximum(r.assignments)
# ...

Custom Splitter

To add a new splitting strategy:

1. Subtype SplitStrategy.
2. Implement a core function (e.g., mysplit(N, s, rng, data)) that returns (train_pos, test_pos) for positions 1:N.
3. Implement _split(data, s; rng) to call split_with_positions(data, s, mysplit; rng=rng).
4. Use ValidFraction for fraction validation.

Example:

struct MySplit <: SplitStrategy; frac; end

function mysplit(N, s, rng, data)
    cut = floor(Int, s.frac * N)
    train_pos = 1:cut
    test_pos = (cut+1):N
    return train_pos, test_pos
end

function _split(data, s::MySplit; rng=Random.default_rng())
    split_with_positions(data, s, mysplit; rng=rng)
end