Three Learning Paradigms

flowchart TD
    ML["Machine Learning"]
    SL["Supervised Learning\n🏷️ Labelled data\nClassification · Regression"]
    UL["Unsupervised Learning\n🔍 No labels\nClustering · Dimensionality reduction"]
    RL["Reinforcement Learning\n🎮 Trial and error\nAgent · Environment · Reward"]
    ML --> SL
    ML --> UL
    ML --> RL

The tree main types of machine learning.

Supervised Learning

The model is trained on input–output pairs. Given input \(x\), it learns to predict output \(y\).

• Classification: predict a category (spam/not spam, species, disease/healthy)
  
• Regression: predict a number (house price, temperature)

from sklearn.neighbors import KNeighborsClassifier

clf = KNeighborsClassifier(n_neighbors=5)
clf.fit(X_train, y_train)
accuracy = clf.score(X_test, y_test)

x_min, x_max = X["petal length (cm)"].min() - 0.5, X["petal length (cm)"].max() + 0.5
y_min, y_max = X["petal width (cm)"].min() - 0.5, X["petal width (cm)"].max() + 0.5
xx, yy = np.meshgrid(np.linspace(x_min, x_max, 300),
                     np.linspace(y_min, y_max, 300))
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
Z_num = np.array(
    [{"setosa": 0, "versicolor": 1, "virginica": 2}[z] for z in Z]
).reshape(xx.shape)

fig, ax = plt.subplots(figsize=(7, 5))
ax.contourf(xx, yy, Z_num, alpha=0.15, cmap="coolwarm")
for species, colour in colours.items():
    mask = y == species
    ax.scatter(X.loc[mask, "petal length (cm)"], X.loc[mask, "petal width (cm)"],
               label=species, color=colour, edgecolors="white", s=60, alpha=0.9)
ax.set_xlabel("Petal length (cm)")
ax.set_ylabel("Petal width (cm)")
ax.set_title(f"Supervised learning — KNN classifier")
ax.legend(title="Species")
plt.tight_layout()
plt.show()

KNN classifier on Iris. The shaded regions are the model’s decision boundaries — areas where it predicts each class. These boundaries were learned entirely from the training data.

Note

The shaded regions are the model’s decision boundaries — areas where it predicts each class. These boundaries were learned entirely from the training data.

---

Unsupervised Learning

There are no labels. The model explores raw data to find hidden structure — groupings, patterns, or compressed representations.

“Can you find natural groups in this data without being told what they are?”

from sklearn.cluster import KMeans

X_unlabelled = df[["petal length (cm)", "petal width (cm)"]]
kmeans = KMeans(n_clusters=3, random_state=42, n_init="auto")
cluster_labels = kmeans.fit_predict(X_unlabelled)

fig, axes = plt.subplots(1, 2, figsize=(12, 4))

axes[0].scatter(X_unlabelled["petal length (cm)"], X_unlabelled["petal width (cm)"],
                color="grey", alpha=0.6, edgecolors="white", s=60)
axes[0].set_title("Raw data — no labels provided")
axes[0].set_xlabel("Petal length (cm)")
axes[0].set_ylabel("Petal width (cm)")

cluster_colours = ["#4C72B0", "#55A868", "#C44E52"]
cluster_names   = ["Cluster A", "Cluster B", "Cluster C"]
for i, (colour, name) in enumerate(zip(cluster_colours, cluster_names)):
    mask = cluster_labels == i
    axes[1].scatter(X_unlabelled.loc[mask, "petal length (cm)"],
                    X_unlabelled.loc[mask, "petal width (cm)"],
                    label=name, color=colour, alpha=0.8, edgecolors="white", s=60)
axes[1].scatter(kmeans.cluster_centers_[:, 0], kmeans.cluster_centers_[:, 1],
                marker="X", s=200, color="black", zorder=5, label="Centroids")
axes[1].set_title("K-Means — structure discovered automatically")
axes[1].set_xlabel("Petal length (cm)")
axes[1].legend()
plt.tight_layout()
plt.show()

K-Means clustering discovers structure without labels

Other unsupervised techniques:

• Dimensionality reduction (PCA, t-SNE): compress many features into 2–3 for visualisation
  
• Anomaly detection: find unusual data points (fraud, manufacturing defects)
  
• Topic modelling: discover themes in large text collections

---

Reinforcement Learning

An agent learns by interacting with an environment. It receives a reward for good actions and a penalty for bad ones — no labelled dataset required.

flowchart LR
    A["🤖 Agent"] -->|"Action"| E["🌍 Environment"]
    E -->|"Reward / Penalty"| A

The reinforcement learning loop

Component	Example: chess
Agent	The chess-playing program
Environment	The board and rules
State	Current board position
Action	Choosing which piece to move
Reward	+1 win, −1 loss

# Illustrative simulation — not a real RL environment
np.random.seed(7)
episodes = np.arange(1, 501)
reward = np.cumsum(
    0.3 + 0.004 * episodes + np.random.normal(0, 2, len(episodes))
)

fig, ax = plt.subplots(figsize=(8, 4))
ax.plot(episodes, reward, color="#4C72B0", linewidth=1.5, alpha=0.85)
ax.axhline(0, color="grey", linestyle="--", linewidth=0.8)
ax.fill_between(episodes, reward, alpha=0.1, color="#4C72B0")
ax.set_xlabel("Episode (one full interaction with the environment)")
ax.set_ylabel("Cumulative reward")
ax.set_title("Reinforcement learning — agent improves through trial and error")
plt.tight_layout()
plt.show()

Simulated RL learning curve: cumulative reward over episodes

Famous RL successes:

• AlphaGo / AlphaZero — mastered Go, chess, and shogi from self-play alone
  
• OpenAI Five — defeated world-champion Dota 2 teams
  
• Robotics — teaching robot arms to grasp and manipulate objects
  
• RLHF — Reinforcement Learning from Human Feedback; used to align large language models