What Is Machine Learning?

First lesson of the Introduction to Machine Learning (scikit-learn) course, free to read: about 15 minutes, 3 exercises and 3 quiz questions. This is the catalogue's base lesson; in the student space, it is rewritten for your profile.

Objectives

Distinguish supervised from unsupervised learning
Identify the features (X) and the target (y) in a dataset
Recognize a regression problem and a classification problem

Machine learning Features and target variable Types of learning

In classical programming, you write the rules: “if the amount exceeds €1,000 and the country is unusual, then flag the transaction”. In machine learning, you provide examples and an algorithm learns the rule from the data. It is useful when the rules are too numerous, too fuzzy, or change over time.

Basic vocabulary

Sample (or observation): a row of the table, for example a customer or a flower.
Feature (explanatory variable): a column used to predict, for example age or the length of a petal.
Target (label): what you want to predict, for example “has churned” or “price”.
Model: the learned function that links the features to the target.

In scikit-learn, X denotes the feature matrix, of shape (n_samples, n_features), and y the target vector, of shape (n_samples,).

Supervised or unsupervised?

Supervised: each example has a known target. The model learns to predict it for new examples.
- Classification: the target is a category (spam / not spam, flower species).
- Regression: the target is a continuous number (apartment price, revenue).
Unsupervised: no target. You look for structure in the data: groups of similar customers (clustering), compressing information (dimensionality reduction).

A first dataset

scikit-learn provides built-in datasets, available without an internet connection. The most famous is Iris: 150 flowers described by 4 measurements, split into 3 species.

from sklearn.datasets import load_iris

iris = load_iris()
X = iris.data          # 2D numpy array: 150 rows, 4 columns
y = iris.target        # 1D numpy array: 150 values (0, 1, or 2)

print(X.shape)               # (150, 4)
print(y.shape)               # (150,)
print(iris.feature_names)    # names of the 4 features
print(iris.target_names)     # ['setosa' 'versicolor' 'virginica']

You can also get a pandas DataFrame, which is handier for exploring:

iris_df = load_iris(as_frame=True)
df = iris_df.frame               # features + 'target' column
print(df.head())
X = df.drop(columns="target")   # X: 2D DataFrame
y = df["target"]                # y: 1D Series

Here the target is a species: so this is a supervised classification problem with 3 classes.

What to remember

Before writing a single line of model code, ask yourself three questions: what is my target? is it categorical or numeric? which features do I have available at prediction time? This last question is essential: a feature that exists only after the event to predict (for example the cancellation date to predict cancellation) completely skews the model. This is called data leakage, and we will come back to it throughout the track.

Exercises

Solutions and tests stay on the server. In the student space, your code is run and checked, and the tutor gives hints before the solution.

CodeDifficulty 1Python in the browser2 hints

Extracting X and y from the Iris dataset

Load the Iris dataset with load_iris(as_frame=True). Create X, a DataFrame containing only the 4 features, and y, the Series of the target column. Then store the number of samples in n_echantillons and the number of features in n_features.

from sklearn.datasets import load_iris

df = load_iris(as_frame=True).frame
X = ...
y = ...
n_echantillons = ...
n_features = ...

CodeDifficulty 1Python in the browser2 hints

Classifying problems

Complete the types dictionary by associating each problem with the value 'classification', 'regression', or 'non_supervise'.

types = {
    "prix_appartement": ...,        # prédire le prix en euros
    "email_spam": ...,              # spam ou non spam
    "groupes_clients": ...,         # trouver des groupes sans étiquette
    "temperature_demain": ...,      # prédire une température
    "espece_fleur": ...,            # setosa, versicolor ou virginica
}

Multiple choiceDifficulty 11 hint

Shape of X

In scikit-learn, what shape should the matrix X have for 200 customers described by 5 variables?

(200, 5)
(5, 200)
(200,)
(1000,)

End-of-lesson quiz

Predicting whether a customer will cancel their subscription (yes/no) is a problem of:

Regression
Classification
Clustering
Dimensionality reduction

What is the main difference between supervised and unsupervised learning?

Supervised learning uses more data
Unsupervised learning has no target variable
Supervised learning uses no features
Unsupervised learning is always more accurate

Why should the ‘cancellation date’ not be used as a feature to predict cancellation?

Because dates are forbidden in scikit-learn
Because it is data leakage: the information exists only after the event
Because it takes too much memory
Because it is categorical