📘 Midterm Review — (Chapters 1–4)

Chapter 1: The Machine Learning Landscape

Important Concepts

• Types of ML:
  • Supervised (e.g., spam detection)
  • Unsupervised (e.g., clustering)
  • Reinforcement (e.g., training a robot)
• Supervised vs Unsupervised:
  • Supervised: Training data has labels.
  • Unsupervised: No labels, only input features.
• Reinforcement Learning:
  • Learn by interacting with environment and getting rewards.
• Batch vs Online Learning:
  • Batch: Train once on full data.
  • Online: Train continuously with incoming data.
• Concept Drift:
  • When the data distribution changes over time.
• Instance-based vs Model-based Learning:
  • Instance-based: Memorizes examples.
  • Model-based: Learns a general model.
• Regression vs Classification:
  • Regression predicts numbers (e.g., house prices).
  • Classification predicts categories (e.g., cat vs dog).
• Challenges in ML:
  • Insufficient data
  • Poor quality data
  • Overfitting / Underfitting

Chapter 2: End-to-End Machine Learning Project

Key Topics

• Define objective early:
  • To align solution with business goals.
• ML Project Steps:
  • Frame the problem → Get data → Explore → Prepare → Model → Evaluate → Deploy.
• Importance of Test Set Early:
  • Prevents data leakage.
• Stratified Sampling:
  • Keep same proportions of important classes.
• Data Exploration Checklist:
  • Name
  • Type
  • Missing values %
  • Noisiness
  • Usefulness
• Handle Missing Values:
  • Drop rows
  • Drop columns
  • Impute (fill)
• Feature Engineering:
  • Create new features to improve performance.