MLFlow End to End Tutorial

---

What is MLFlow?

---

• MLflow is opensource platform to Manage the ML Life Cycle.
• It includes experimentation, reproducibility, deployment and central model registry.
• Advantage of Use MLFlow is Transparency and Standardization.
• It helps you to train, reuse and deploy the model.

MLFlow Setup

---

1. Installation of MLflow.

pip install mlflow

2. Once you will install with this now time to setup Central Repository Database where we will log all our tracking information and also will create the artifacts folder to store our models and its relevant information.

# Terminal Command
mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./artifacts --host 0.0.0.0 

or

mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./artifacts --host 127.0.0.1 --port 5000

Output

This Command create the mlflow.db and artifacts folder.

Open MLFLOW UI

---

mlflow ui

MLFLOW UI LOOKS LIKE

---

MLflow Tracking

---

• MLflow Tracking is an API and UI for Logging Parameter, Code Versions, Metrics, and Output files.

Set Tracking URI and Create Experiment

• We have to create the notebook for tracking server will run in. and we have to set the tracking URI. In our example, localost:5000 is the tracking URI but you can setup any other tracking host and port in the set_tracking_uri function call.
• Initially we have to default experiment created which you can get with get_experiment function call.

⭐ mlflow.py

---

experiment_id = mlflow.create_experiment("my_experiment")
experiment = mlflow.get_experiment(experiment_id)

mlflow.set_tracking_uri("http://localhost:5000")
experiment = mlflow.get_experiment('0')
print("Name: {}".format(experiment.name))
print("Artifact Location: {}".format(experiment.artifact_location))
print("Life Cycle Stage: {}".format(experiment.lifecycle_stage))
print("Experiement ID: {}".format(experiment.experiment_id))

🤝 Prepare the Notebook with ML Case study

---

For Experiment Run Notebook, we have to create the notebook with the following steps.

Notebook	Link
Notebook	Open

Notebook Flow

---

MLFlow Model Flavors

---

The model can be used with tools that support either the sklearn or python_function model flavors.

Diverse Platform

---

MLflow provides tools to deploy many common model types to diverse platforms.

Experiments

---

Final Model Prediction from Staging Code

---

# Predict with MLflow Model
print("Predict with MLflow Model:")
# models:///XGBoost_Model/{Stage} -- Stage Contains Two Stage 1. Staging 2. Production(Current)
model = mlflow.xgboost.load_model("models:///XGBoost_Model/Production")
print("="*50)
print("Model:\n", model)
print("="*50)
prediction = model.predict(X_test)
# print("Prediction.type:", type(prediction))
# print("="*50)
# print("Prediction.shape:", prediction.shape)
# print("="*50)
print("Prediction:\n", prediction)
print("Prediction Done.")

---

Thanks for Reading 👨‍💻👩‍💻🧑‍💻