MLflow Model Registry in MLOps - Step-by-Step Execution

Process Flow - MLflow Model Registry

Register Model

↓

Create Model Version

↓

Transition Stage

↓

Staging

↓

Archive Old Versions

↓

End

This flow shows how a model is registered, versioned, moved through stages like staging and production, and old versions archived.

Execution Sample

MLOps

import mlflow

# Register model
model_uri = "runs:/12345/model"
model_details = mlflow.register_model(model_uri, "MyModel")

# Transition model stage
client = mlflow.MlflowClient()
client.transition_model_version_stage("MyModel", model_details.version, "Staging")

This code registers a model from a run and moves it to the Staging stage in the MLflow Model Registry.

Process Table

Step	Action	Input	Output/Result
1	Register model	model_uri = runs:/12345/model	Model 'MyModel' version 1 created
2	Create MLflow client	None	Client object ready
3	Transition model stage	Model 'MyModel' version 1 to 'Staging'	Model version 1 stage updated to 'Staging'
4	Check model stage	Query model version 1	Stage = 'Staging'
5	Archive old versions	None	No old versions to archive
6	End	None	Process complete

💡 All steps completed successfully; model is now in Staging stage.

Status Tracker

Variable	Start	After Step 1	After Step 3	Final
model_uri	None	runs:/12345/model	runs:/12345/model	runs:/12345/model
model_details.version	None	1	1	1
model_stage	None	None	Staging	Staging

Key Moments - 2 Insights

Why do we need to create a client object before transitioning the model stage?

The client object (mlflow.MlflowClient) is required to interact with the Model Registry API, as shown in step 2 and used in step 3 to change the model stage.

What happens if we try to transition a model version that does not exist?

An error will occur because the model version must exist before changing its stage. Step 1 ensures the version is created before step 3 transitions it.

Visual Quiz - 3 Questions

Look at the execution table, what is the model stage after step 3?

A Staging

B Production

C Archived

D None

💡 Hint

Check the 'Output/Result' column in step 3 and step 4 for the model stage.

At which step is the model version created?

A Step 3

B Step 2

C Step 1

D Step 5

💡 Hint

Look at the 'Action' and 'Output/Result' columns to find when the model version is created.

If we skip creating the client object, what will happen at step 3?

A Model stage will transition successfully

B An error will occur because client is undefined

C Model version will be deleted

D Nothing will happen

💡 Hint

Step 3 uses the client object to transition the model stage; without it, the code cannot run.

Concept Snapshot

MLflow Model Registry manages model versions and stages.
Use mlflow.register_model() to register a model.
Use MlflowClient.transition_model_version_stage() to move stages.
Stages include None, Staging, Production, Archived.
Keep track of versions to deploy or archive models safely.

Full Transcript

This visual execution shows how to register a model in MLflow Model Registry, create a version, and transition it to the Staging stage. First, the model URI from a run is registered, creating version 1. Then, a client object is created to interact with the registry API. Using this client, the model version's stage is changed to Staging. The execution table tracks each step's input and output, confirming the model stage update. Variables like model_uri, model_details.version, and model_stage are tracked through the process. Key moments clarify why the client is needed and what happens if the version does not exist. The quiz tests understanding of model stages, version creation, and client usage. The snapshot summarizes the main commands and concepts for quick reference.