Clustering with KMeans: Practice Exercise

What you'll learn

• How to implement KMeans clustering on simple and complex datasets
• How to evaluate clustering results using centroids and inertia
• How to visualize clusters using matplotlib
• How to predict cluster membership for new data points
• How to analyze and filter data by cluster assignments

:computer: Activity Type: Hands-on coding exercise ⏱️ Total Time: 45-60 minutes

Getting Started

Access the exercise notebook here: KMeans Clustering Colab Notebook

:bulb: Before you begin: Make sure you're signed in to your Google account to save a copy of the notebook!

Exercise Overview

Part One: KMeans with a Simple Dataset

⏱️ Time: 15-20 minutes

In this part, you will practice basic KMeans clustering on a small dataset.

Step-by-step instructions:

1. Import the necessary libraries
   • numpy for data handling
   • sklearn.cluster for KMeans
   • matplotlib.pyplot for visualization
2. Create and explore the sample dataset
   • Review the provided data points
   • Understand the 2D structure
3. Apply KMeans clustering
   • Initialize KMeans with n_clusters=2
   • Set random_state=42 for reproducibility
   • Fit the model to your data
4. Evaluate the model
   • Check the cluster centroids
   • Calculate the inertia (sum of squared distances)
5. Make predictions
   • Predict the cluster for point [10, 5]
   • Understand which cluster it belongs to
6. Visualize the results
   • Plot the data points colored by cluster
   • Mark the centroids clearly
   • Add the predicted point tip Visualization tip: Use different colors for each cluster and mark centroids with a special symbol (like 'X' or '*') to make them stand out!

---

Part 2: KMeans with a Larger Dataset

⏱️ Time: 25-30 minutes

Now you'll work with a real-world dataset to see how clustering performs on complex data.

Step-by-step instructions:

1. Download and load the dataset

   • Use gdown to download Live.csv
   • Load it using pandas
   • Display the first few rows

2. Explore the dataset

   • Check the shape and columns
   • Look for missing values
   • Understand what each feature represents

3. Select features for clustering

   • Choose numerical features only
   • Consider which features are most relevant
   • Create a feature matrix

4. Apply KMeans clustering

   • Initialize with n_clusters=4
   • Use random_state=42 for consistency
   • Fit the model to your selected features

5. Analyze the results

   • Examine the cluster centroids
   • Check the inertia value
   • Add cluster labels to your dataframe

6. Make a prediction

   • Predict cluster for: [529, 200, 80, 430, 100, 20, 50, 6, 3]
   • Interpret what this cluster assignment means

7. Filter and analyze by cluster

   • Create subsets for each cluster
   • Calculate summary statistics per cluster
   • Look for patterns

8. Visualize the clusters

   • Choose 2 features for 2D plotting
   • Color points by cluster
   • Add centroids to the plot

:bulb: Data tip: If you have many features, consider using the first two principal components for visualization, or choose the two most meaningful features for your analysis.

:warning: Warning Common challenge: Make sure your feature array matches the order of features used during training when making predictions!

Part 3: Advanced Challenges (Optional)

⏱️ Time: 10-15 minutes

For those who finish early or want extra practice:

1. Optimal cluster selection
   • Implement the elbow method
   • Plot inertia vs. number of clusters
   • Find the optimal n_clusters
2. Dimensionality reduction
   • Apply PCA before clustering
   • Compare results with and without PCA
   • Visualize in the reduced space
3. Alternative clustering
   • Try different initialization methods
   • Experiment with MiniBatchKMeans for larger datasets tip Elbow method hint: Look for the "elbow" point where adding more clusters doesn't significantly reduce inertia.

Exercise Submission

:warning: Warning Important: Save your completed notebook before submitting!

Please submit your work through this link: Exercise Submission Form

Submission Guidelines

1. Complete all required sections

   • Part One: Simple dataset clustering
   • Part 2: Complex dataset analysis
   • Optional: Advanced challenges (bonus points!)

2. Verify your code

   • Ensure all cells run without errors
   • Check that visualizations display correctly
   • Confirm predictions are calculated

3. Use consistent parameters

   • Always use random_state=42
   • Follow the specified n_clusters values
   • Keep variable names consistent

4. Submit your work

   • Download your completed notebook (.ipynb file)
   • Upload it through the submission form
   • Include any additional observations or insights

:information_source: Submission deadline: Check with your instructor for the due date. Late submissions may not receive full credit.

Troubleshooting

Common issues and solutions:

1. Import errors

   • Run !pip install scikit-learn matplotlib pandas if needed
   • Restart runtime after installation

2. Data loading issues

   • Ensure you're connected to the internet
   • Check that gdown is installed: !pip install gdown

3. Visualization problems

   • Use plt.figure() before plotting
   • Call plt.show() to display plots

4. Prediction errors

   • Ensure input shape matches training data
   • Use double brackets for single predictions: [[10, 5]]

:bulb: Tip Need help? Post questions in the course discussion forum or attend office hours!