Skip to content

Commit

Permalink
chap mlflow
Browse files Browse the repository at this point in the history
  • Loading branch information
avouacr committed Feb 16, 2024
1 parent 6d71565 commit 1e87d25
Show file tree
Hide file tree
Showing 6 changed files with 87 additions and 0 deletions.
Binary file added img/mlflow-model-registry.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added img/mlflow-models.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added img/mlflow-overview.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added img/mlflow-tracking.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added img/mlops.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
87 changes: 87 additions & 0 deletions index.qmd
Original file line number Diff line number Diff line change
Expand Up @@ -910,3 +910,90 @@ Source : [k21academy.com](https://k21academy.com/docker-kubernetes/docker-and-ku

# MLOps

## Motivation

- Intégrer :
- les [**principes DevOps**]{.orange}
- les [**spécificités**]{.orange} des projets de ML

. . .

![](img/mlops.png){fig-align="center" height=400}

## MLOps : principes

- [**Reproductibilité**]{.orange}

- [**Contrôle de version**]{.orange}

- [**Automatisation**]{.orange}

- [**Collaboration**]{.orange}

- [***Monitoring***]{.orange}

## MLOps : implémentation

- De nombreux [**frameworks**]{.orange} implémentent les principes du MLOps
- Catalogue du `SSP Cloud` : [MLFlow](https://mlflow.org/)

- Avantages de `MLflow` :
- [**Open-source**]{.blue2}
- Couvre l'entièreté du [**cycle de vie**]{.blue2} d'un modèle ML
- [**Agnostique**]{.blue2} au package ML utilisé
- Intégration avec `Kubernetes`

## `MLFlow` : vue d'ensemble

![](img/mlflow-overview.png){fig-align="center" height=350}

[Source](https://dzlab.github.io/ml/2020/07/12/ml-ci-mlflow/)

## `MLFlow` : Tracking server

- "Une [**API**]{.orange} et une [**interface utilisateur**]{.orange} pour [**enregistrer**]{.orange} les paramètres, les versions du code, les métriques et les artefacts"

. . .

![](img/mlflow-tracking.png){fig-align="center" height=400}

## `MLFlow` : Models

- "Une convention pour [**'packager'**]{.orange} des [**modèles**]{.orange} de *machine learning* sous plusieurs [**formes**]{.orange}"

. . .

![](img/mlflow-models.png){fig-align="center" height=400}

## `MLFlow` : Model registry

- "Un [**entrepôt centralisé de modèles**]{.orange}, un ensemble d'API et une interface utilisateur pour gérer [**collaborativement**]{.orange} le cycle de vie complet d'un modèle MLflow"

. . .

![](img/mlflow-model-registry.png){fig-align="center" height=400}

## Servir le modèle (sans `MLFlow`)

## Servir le modèle (avec `MLFlow`)






# Application

## MLOps

- Consignes sur le [site du cours](https://ensae-reproductibilite.github.io/website/chapters/application.html)
- Partie X : application des principes MLOps avec `MLFlow`






# Conclusion

Vision essentiellement technique dans ce cours mais changements plus larges (organisationnels, etc..)

0 comments on commit 1e87d25

Please sign in to comment.