This is an old revision of the document!
Hands-on Lab 3: Build a Complete ML Pipeline Using AutoML and Designer 🧭 Objective In this lab, you will:
Use Azure Machine Learning Designer for a visual ML pipeline
Build and evaluate a model using AutoML
Compare both methods
Deploy the best model
🧱 Prerequisites An active Azure subscription
Azure Machine Learning Workspace already created
Compute instance and compute cluster set up
A basic dataset (e.g., Titanic, Diabetes, or Bank Marketing)
⚙️ PART 1: Create a Dataset in Azure ML Go to Azure ML Studio
Navigate to “Data” > + Create
Choose “From local files” or “Web URL” and upload your dataset (e.g., diabetes.csv)
Set type: Tabular
Choose your default datastore
Confirm schema and create the dataset
🧪 PART 2: AutoML Pipeline 🧩 Step 1: Launch AutoML In the Azure ML Studio, go to “Automated ML”
Click + New Automated ML run
Select your uploaded dataset
Select or create a new experiment
Choose a compute cluster
🧪 Step 2: Configure AutoML Task type: Choose (e.g., Classification if using Titanic)
Target column: e.g., Survived
Accept default preprocessing
Set Exit criteria (e.g., 1 hour or 20 iterations)
▶️ Step 3: Start Run Click Start
Wait for the experiment to finish (AutoML will try many algorithms and preprocessing pipelines)
🧠 PART 3: Build the Same Model Using Designer 📊 Step 1: Open Designer In Azure ML Studio, go to “Designer”
Click + New pipeline
📦 Step 2: Drag Components From the left panel, drag the following:
Dataset (your uploaded dataset)
Select Columns in Dataset
Clean Missing Data (optional)
Split Data
Train Model
Score Model
Evaluate Model
Connect components in this order:
mathematica Copy Edit Dataset → Select Columns → Clean Missing → Split Split → Train Model (also needs Label column) Train → Score → Evaluate Choose a built-in algorithm from “Model” tab, e.g., Logistic Regression.
🛠 Step 3: Configure Parameters In Train Model, choose the label column
In Split Data, use 0.7/0.3 ratio
Select scoring metric: e.g., accuracy
▶️ Step 4: Submit Job Select compute target
Click Submit
Wait for job to finish
View metrics in Evaluate Model
🚀 PART 4: Register and Deploy the Best Model From AutoML results or Designer output, click “Register Model”
Go to “Models” tab in left nav
Click your registered model → Deploy
Choose Real-time endpoint
Select compute type (e.g., Azure Kubernetes Service or Managed Online Endpoint)
Add sample input data for testing
📈 PART 5: Compare & Document Method Accuracy Time Model Name AutoML 0.89 25m AutoML_001 Designer 0.85 10m Logistic_001
📁 Deliverables ✅ A completed visual pipeline in Designer ✅ An AutoML experiment with leaderboard ✅ A registered and deployed model ✅ A short report comparing the two methods
🧼 Cleanup Delete compute cluster if no longer needed
Delete endpoints to avoid charges
