This quickstart tutorial demonstrates how to use Azure Machine Learning to train a simple linear regression model on the Boston House Prices dataset.

aml_quickstart_tutorial/
├── README.md                 # This file
├── requirements.txt          # Python dependencies
├── .env.template            # Environment variables template
├── data/
│   └── boston_house_prices.csv  # Sample dataset
├── code/
│   ├── train_model.py       # Model training script
│   └── upload_data_asset.py # Script to upload data to Azure ML
└── jobs/
    └── submit_training_job.py   # Script to submit training job to Azure ML

1. An Azure subscription
2. An Azure Machine Learning workspace
3. Python 3.8 or later
4. Azure CLI (optional, for additional commands)

To set up the environment automatically run:

cd aml_quickstart_tutorial
./setup.sh

or you can also do it step by step:

cd aml_quickstart_tutorial
pip install -r requirements.txt

1. Copy the environment template:

   cp .env.template .env

2. Edit .env and fill in your Azure details:

   AZURE_SUBSCRIPTION_ID=your_subscription_id_here
   AZURE_RESOURCE_GROUP=your_resource_group_here
   AZURE_ML_WORKSPACE_NAME=your_workspace_name_here
   AZURE_ML_COMPUTE_NAME=cpu_cluster_name_here
   

Make sure you're authenticated with Azure. You can use one of these methods:

• Azure CLI: az login
• VS Code: Use the Azure extension
• Environment variables: Set AZURE_CLIENT_ID, AZURE_CLIENT_SECRET, AZURE_TENANT_ID

Before running on Azure ML, you can test the training script locally:

cd code
python train_model.py --data_path ../data/boston_house_prices.csv --output_dir ./local_outputs

Upload the dataset as a data asset to your Azure ML workspace:

cd code
python upload_data_asset.py

This creates a data asset named "boston-house-prices" that can be reused across experiments.

Submit the training job to run on Azure ML compute:

cd jobs
python submit_training_job.py

This will:

• Create a compute environment with the required dependencies
• Submit a training job using the uploaded data asset
• Train a linear regression model
• Save the trained model and metrics

After submitting the job, you can monitor it:

1. Azure ML Studio: Use the Studio URL provided in the output
2. Azure CLI: az ml job show --name <job_name>
3. Stream logs: az ml job stream --name <job_name>

The training script (train_model.py):

1. Loads the Boston House Prices dataset
2. Splits the data into training and test sets
3. Trains a linear regression model using scikit-learn
4. Evaluates the model and logs metrics to Azure ML
5. Saves the trained model and feature names

• Data Asset Management: The dataset is uploaded as a versioned data asset
• Environment Management: Automatic creation of conda environment with dependencies
• Experiment Tracking: Metrics are logged to Azure ML for comparison
• Model Artifacts: Trained models are saved and can be registered for deployment
• Scalable Compute: Jobs run on Azure ML compute clusters

The training script logs the following metrics:

• R² Score: Coefficient of determination (how well the model explains variance)
• RMSE: Root Mean Square Error
• MSE: Mean Square Error

After completing this quickstart, you can:

1. Register the Model: Register the trained model for deployment
2. Create Endpoints: Deploy the model as a real-time or batch endpoint
3. Experiment with Different Models: Try other algorithms like Random Forest
4. Add Data Drift Monitoring: Monitor your deployed model for data drift
5. Set up MLOps Pipelines: Automate training and deployment with Azure DevOps or GitHub Actions

1. Authentication Errors: Make sure you're logged in with az login
2. Compute Not Found: Ensure your compute cluster exists or update the compute name in .env
3. Environment Creation Fails: Check that all dependencies in requirements.txt are available
4. Data Asset Not Found: Make sure you've run the upload script first

• Azure ML Documentation
• Azure ML Python SDK
• Azure ML CLI

The Boston House Prices dataset contains:

• Features: 13 attributes including crime rate, property tax, pupil-teacher ratio, etc.
• Target: Median home value (MEDV)
• Samples: ~500 housing records
• Task: Regression (predicting continuous values)

This is a classic dataset for learning regression techniques, though note that it's considered outdated for real-world applications due to ethical concerns with some features.