ML Made Simple

Exercise Part 4: Train a Machine Learning Model Using Azure ML

Ready to play the role of a machine learning maestro? In this exercise, we’ll dive into Microsoft Azure Machine Learning’s automated feature to train a model that predicts bicycle rentals. Think of it as a chef trying different ingredients to whip up the best dish!

---

Step 1: Understanding Automated Machine Learning

• What It Does: Automated machine learning tests various methods and algorithms to find the best model for your data. It’s like a talent show for algorithms, where the best performer wins!

---

Step 2: Setting Up an Experiment

1. Automated ML Page: In Azure Machine Learning studio, find the Automated ML page (look under Author).
2. Create a New Automated ML Run:
   • Select Dataset: Use the ‘bike-rentals’ dataset.
   • Configure Run:
     • Experiment Name: mslearn-bike-rental
     • Target Column: rentals (this is the main star, the value you want to predict)
     • Training Compute Target: Use the compute cluster you created earlier.
   • Task Type and Settings:
     • Task Type: Regression (since you’re predicting a number)
     • Additional Configurations:
       • Primary Metric: Normalized root mean square error (a fancy way of measuring accuracy)
       • Explain Best Model: Selected (to understand which features matter most)
       • Blocked Algorithms: Only use RandomForest and LightGBM (to save time)
   • Exit Criterion:
     • Training Job Time (hours): 0.25 (15 minutes)
     • Metric Score Threshold: 0.08 (stops the experiment if it finds a model this good or better)
   • Featurization Settings:
     • Enable Featurization: Selected (let Azure ML prep the data for you)

---

Step 3: Running the Experiment

• Start the Magic: Submit the details and watch as the experiment kicks off. First, it will prepare, then start running. It might be coffee break time while you wait!

---

Step 4: Reviewing the Best Model

1. Best Model Summary: Once the experiment is done, check out the Details tab for a summary of the winning model.
2. Model Details: Select the Algorithm name to dive deeper. The model’s success is based on how small the Normalized root mean square error is – the smaller, the better.

---

Step 5: Analyzing Model Performance

1. Check Other Metrics: On the Explanations tab, explore other evaluation metrics for your regression model.
2. Understanding Predictions and Errors:
   • Predicted vs. True Chart: Look for a diagonal trend showing that predictions are close to the actual values.
   • Residual Histogram: Here, you want most errors (differences between predicted and actual values) to be small, clustered around 0.

---

Step 6: Feature Influence

• Global Importance Chart: This shows which features in your dataset most influence the label prediction. It’s like finding out which spices made the dish a hit!

---

And there you have it! You’ve just trained a machine learning model to predict bicycle rentals. Who knew numbers could tell such exciting stories?