Curriculum/combinatorially-symmetric-cv
Combinatorially Symmetric CV
research process·L3 · system pattern·stub
Replacesthe belief that one train/test split is enough.
CPCV (López de Prado, AFML ch. 12) generalises purged k-fold to all symmetric combinations of train/test groups — for k=10 groups, that's `C(10, k/2) = 252` distinct splits. The aggregate over splits gives a distribution of out-of-sample performance, not a single number. The distribution shape — particularly the rank correlation between in-sample and out-of-sample performance across splits — is what PBO measures.
Prerequisites
Bridges
- monte-carlo-vs-cpcvshared mechanismMonte Carlo CV samples random train/test splits. CPCV enumerates all combinatorially symmetric ones. For the same k, CPCV gives more splits, more stable distribution estimates, and deterministic results — at the cost of compute time. Trade is usually worth it for trustworthiness.
This concept is a node in the curriculum DAG. The full lab — page blocks, done state, references — has not been authored yet. The relations above describe where it sits in the graph.
Author at: content/concepts/combinatorially-symmetric-cv/card.ts