This Jupyter notebook demonstrates the implementation of K-Means clustering on a U.S. crime dataset using Amazon SageMaker's built-in KMeans algorithm. Here’s what the notebook does step-by-step:

---

1. Data Acquisition

• The notebook downloads a CSV dataset (crime.csv) containing crime statistics for different U.S. states.

---

2. Data Preparation

• Loads the dataset into a pandas DataFrame (crime).
• Defines a function, stateToNumber, to convert state names into unique numeric codes (by summing hex values of their character codes).
• Replaces the 'State' column in the DataFrame with these numeric codes.
• Converts the DataFrame to a Numpy array of type float32 for processing.

---

3. Feature Selection

• Slices the data to select relevant columns for clustering (columns 1-4, typically those representing different crime statistics).

---

4. Setting up Amazon SageMaker for Clustering

• Imports SageMaker modules.
• Acquires the AWS execution role and sets up S3 bucket paths for data input and output.
• Initializes a SageMaker KMeans estimator, configuring parameters including:
  • Number of clusters (k=10)
  • Instance type and count for training
  • Input and output S3 locations

---

5. Training the KMeans Model

• Trains the KMeans model on the selected crime data slice using SageMaker’s distributed training infrastructure.

---

6. Model Deployment and Prediction

• Deploys the trained KMeans model as a real-time endpoint on SageMaker.
• Supplies state data (crime statistics) to the endpoint predictor for clustering.
• The predictor assigns each state to its closest cluster.

---

7. Results Interpretation

• For each state, the notebook prints:
  • State identifier
  • State code
  • Closest cluster label
  • Crime statistics (murders, assaults, urban population, rapes)
• These outputs allow users to analyze how states are grouped based on their crime profiles.

---

Summary

This notebook is a practical example of:

• Preprocessing tabular crime data (feature engineering and transformation)
• Using Amazon SageMaker for scalable K-Means training and deployment
• Assigning cluster labels to U.S. states according to their crime characteristics
• Presenting clustering results for interpretation and analysis

It covers the full pipeline: download data, clean and process data, train and deploy a model using cloud resources, and visualize results for clustering-based insights.