Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/ageron/handson-ml3/llms.txt

Use this file to discover all available pages before exploring further.

Chapter 1 sets the stage for the entire book by mapping out the machine learning landscape. Rather than diving into code immediately, it builds the conceptual vocabulary you need to understand, compare, and evaluate ML systems. You will learn how to categorize algorithms, assess the quality of a training set, and identify the failure modes that cause models to underperform.

​
What you’ll learn

  • The three major learning paradigms: supervised, unsupervised, and reinforcement learning
  • The difference between batch (offline) learning and online (incremental) learning
  • Instance-based learning versus model-based learning
  • The main challenges of ML: insufficient data, non-representative data, poor-quality data, irrelevant features, overfitting, and underfitting
  • How to frame a machine learning problem and select an appropriate performance measure
  • A first end-to-end example predicting life satisfaction from GDP per capita

​
Key concepts

Types of machine learning. Supervised learning algorithms train on labeled data—each training example pairs an input with a desired output. Classification (predicting a category) and regression (predicting a continuous value) are both supervised tasks. Unsupervised learning finds hidden structure in unlabeled data: clustering groups similar instances together, anomaly detection identifies unusual examples, and dimensionality reduction compresses data while preserving structure. Reinforcement learning trains an agent to maximize cumulative reward through environment interaction. Batch versus online learning. A batch (offline) learning system trains on the full dataset and is then deployed without further updates. When new data arrives the model must be retrained from scratch. Online learning systems, by contrast, update the model incrementally as each new data point—or small mini-batch—arrives. Online learning is well-suited to continuous data streams and situations where retraining from scratch would be prohibitively expensive. Instance-based versus model-based learning. Instance-based learners (such as k-nearest neighbors) generalize by comparing new inputs to stored training examples using a similarity measure. Model-based learners (such as linear regression) fit a parametric function to the training data and then use that function to make predictions on new inputs. Model-based learning is typically faster at inference time because no search over stored examples is required. Overfitting and underfitting. Overfitting occurs when a model captures noise in the training data rather than the true underlying pattern; the model performs well on training examples but poorly on unseen data. Regularization—adding a penalty on model complexity—is the primary tool for preventing overfitting. Underfitting occurs when the model is too simple to capture the structure of the data; the remedy is a more powerful model, better features, or less regularization.

​
Code examples

The first concrete example in Chapter 1 loads life-satisfaction data for several countries and fits a linear model to predict satisfaction from GDP per capita: 
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from sklearn.linear_model import LinearRegression

# Download and prepare the data
data_root = "https://github.com/ageron/data/raw/main/"
lifesat = pd.read_csv(data_root + "lifesat/lifesat.csv")
X = lifesat[["GDP per capita (USD)"]].values
y = lifesat[["Life satisfaction"]].values

# Visualize the data
lifesat.plot(kind='scatter', grid=True,
             x="GDP per capita (USD)", y="Life satisfaction")
plt.axis([23_500, 62_500, 4, 9])
plt.show()

# Select a linear model
model = LinearRegression()

# Train the model
model.fit(X, y)

# Make a prediction for Cyprus
X_new = [[37_655.2]]  # Cyprus' GDP per capita in 2020
print(model.predict(X_new))  # outputs [[6.30165767]]
Swapping the model for k-nearest neighbors requires only two lines to change: 
from sklearn.neighbors import KNeighborsRegressor

model = KNeighborsRegressor(n_neighbors=3)
Chapter 1 is primarily conceptual. The notebook generates the lifesat.csv file used in the example above and produces several of the book’s figures. The real hands-on coding starts in Chapter 2.

​
Running this notebook

1

Open in Colab

Click the badge below or navigate directly to the notebook:Open in Colab
2

Run the setup cells

The notebook checks that Python ≥ 3.7 and Scikit-Learn ≥ 1.0.1 are installed, then sets matplotlib defaults.
3

Work through the examples

Run the cells in order. The notebook generates several figures from the book and produces the lifesat.csv dataset used in Code Example 1-1.

​
Exercises

Chapter 1 includes five exercises at the end of the notebook. They ask you to apply the conceptual framework to new scenarios—for example, classifying an ML task as supervised or unsupervised, and identifying likely challenges with a given dataset.

Build docs developers (and LLMs) love

Get started for freeTalk to us