-
Notifications
You must be signed in to change notification settings - Fork 50
Open
Description
-
Intro
-
1er chapitre : Cas simple MLFlow (scikit learn auto log)
-
2eme chapitre : Cas métier sur APE (class custom)
- Contextualisation / Problématique
- Présentation data
- Class custom / différence autolog
-
3eme chapitre :
- Déploiement local puis FastAPI
- Déploiement Kubernetes (manuel)
- Déploiement Kubernetes automatisé (ArgoCD) @ThomasFaria
-
4eme chapitre : Entrainement/optimisation distribué (argo workflow)
-
5eme chapitre : ML en production
- Quelle organisation ?
- Monitoring / observability : quelles métriques ?
- Métier : training-prod skew, data drift, model drift
- (Software : logging API, resources use, latence, ...)
- Construire la chaîne de traitement des logs
- Implémenter le logging pertinent dans le code applicatif (métriques métiers..) @ThomasFaria
- Envoyer les logs sur S3 @avouacr
- ETL pour construire des parquets partitionnés par date @avouacr
- Dashboard avec Superset/Redash @tomseimandi
-
Aller plus loin
- Réentraînements
Metadata
Metadata
Assignees
Labels
No labels