Machine learning: comment éviter le surapprentissage sur un petit dataset ?

poze 13 days ago51 gadefr

3

J'ai seulement ~2000 exemples étiquetés. Le modèle mémorise l'ensemble d'entraînement. Régularisation, augmentation de données, validation croisée — par où commencer ?

🤖 Dyagnostik IA

Se IA ki fè l. Se pa yon repons — kominote a anba a konfime oswa korije l. Toujou verifye anvan ou fye l.

2 Repons

8
Repons yo aksepte

Use a healthcheck and condition: service_healthy:

db:
  healthcheck:
    test: ["CMD-SHELL", "pg_isready -U postgres"]
    interval: 5s
    retries: 5
app:
  depends_on:
    db:
      condition: service_healthy

Also add app-level connection retries — containers restart.

Konekte pou di lòt moun si li te mache.

answered 13 days ago
7

With ~2000 examples: start with cross-validation (5-fold), strong regularization (dropout / weight decay), and early stopping. Then data augmentation. Prefer a small model or fine-tune a pretrained one rather than training from scratch.

Konekte pou di lòt moun si li te mache.

answered 13 days ago

Sign in and verify your email to post an answer.