Split examples before synthesis and preserve evaluation roles

Aperta
#32 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
35/100
Tipo di issue
Funzionalità
Chiarezza
Abbastanza chiara
Stato di attività
Tranquilla
Stack tecnologico
machine-learning, python

Direzione di ricerca

Start by tracing the learning pipeline paths that perform initial and per-class synthesis, refinement, audits, and reporting; the issue does not name specific files or tests. Check where examples enter prompts and feedback, then identify how split identities and group boundaries are carried through those paths. Done means tests confirm the required example IDs and roles, matched-population audit comparisons, and explicitly exploratory reporting for undersized data.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

correctness

Problem

The current learning pipeline can synthesize initial rules before creating the refinement holdout. This means examples later described as held out may already have influenced initial generation.

Repeated candidate selection on one development split also makes its post-selection precision optimistic.

Direction

  • Split examples before every initial synthesis/refinement path.
  • Keep explicit train, selection/validation, calibration, and final-test roles.
  • Preserve group boundaries such as source document or conversation.
  • Carry split identity through refinement, audit, and reporting.
  • Never silently report training metrics as validation when a split is empty or undersized.

Acceptance

  • Initial and per-class synthesis see train examples only.
  • Selection/threshold choice cannot see calibration or final-test examples.
  • Tests verify example IDs exposed to prompts and feedback.
  • Pre/post audit comparisons use the same named population.
  • Small-data behavior is explicitly exploratory rather than mislabeled as validation.

Non-goals

  • A production certification system.
  • A general cross-validation framework.
Lingua principale
Python
Stelle
34
Fork
9
Merge medio
9g 13h
PR unite (30g)
1

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di KRLabsOrg/rulechef

Tutte le issue di KRLabsOrg/rulechef

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.