Classification

Estimated time: 5 hours

Read Chapter 3: Classification from Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow.

The chapter uses the MNIST handwritten-digit dataset. Although images are often described as unstructured data, for this chapter you can think of each image as a collection of numerical features: each pixel becomes a feature, much like variables such as income or age in a tabular dataset.

Focus on the familiar machine learning workflow:

load data → split data → train → predict → evaluate

The purpose of this reading is to deepen your understanding of classification and how classification models should be evaluated.

First Pass: Get the Big Picture — 45 minutes

Focus on the structure and flow of the chapter rather than technical details.

  • Read all section titles and subheadings.
  • Look at the figures, plots, and code outputs and read their captions.
  • Pay particular attention to the MNIST examples and how pixel grids become feature vectors.
  • Skim the exercise questions at the end.

Goal: Build a mental overview of how classification problems are structured.

Second Pass: Read for Understanding — 3 hours

Read the chapter carefully.

Focus on what each stage of the classification workflow is doing rather than memorizing individual lines of code.

Pay particular attention to:

  • how the dataset is represented;
  • how training and test data are separated;
  • how a classifier learns from labelled examples;
  • how predictions are evaluated;
  • accuracy, precision, and recall;
  • binary, multiclass, and multilabel classification.

If you encounter a difficult concept, make a note and continue reading.

Goal: Understand the overall logic of a classification workflow, from data to evaluation.

Third Pass: Connect Text to Code — 1 hour

Create a new Jupyter notebook.

Using the chapter as your guide, reproduce a simple classifier on the MNIST dataset.

Try to work directly from the book rather than following a separate tutorial.

Goal: Connect the conceptual explanation in the chapter to a working classification model.