Types of Machine Learning

Machine learning comes in a few main types, depending on what kind of data you have.

The Three Main Types

Supervised Learning

In supervised learning every example has a label, the correct answer. It is like a student practising with a past paper that has the answers at the back.

Supervised learning has two kinds of problems:

• Classification: the answer is a category, like "spam" or "not spam", "apple" or "guava"
• Regression: the answer is a number, like a price, a temperature or marks

Here a fruit seller in Karachi trains a model to tell apples from guavas using weight and a colour score (10 = very red):

from sklearn.neighbors import KNeighborsClassifier

# Features: [weight in grams, colour score 1-10]
X = [[150, 8], [170, 9], [130, 7], [120, 2], [110, 3], [100, 2]]
y = ["apple", "apple", "apple", "guava", "guava", "guava"]

model = KNeighborsClassifier(n_neighbors=3)
model.fit(X, y)

print(model.predict([[160, 8], [105, 3]]))

['apple' 'guava']

K-Nearest Neighbors looks at the 3 most similar fruits it has seen and takes a vote. A heavy, red fruit is most like the apples.

Unsupervised Learning

In unsupervised learning there are no labels. The model looks for structure on its own.

The most common task is clustering: putting similar items into groups. A shop owner might want to find types of customers without knowing the types in advance:

from sklearn.cluster import KMeans

# Each customer: [visits per month, average bill in thousand Rs]
customers = [[2, 1], [3, 1.5], [2, 2], [12, 8], [10, 9], [11, 7.5]]

kmeans = KMeans(n_clusters=2, random_state=0, n_init=10)
kmeans.fit(customers)

print(kmeans.labels_)

[1 1 1 0 0 0]

The model found two groups: the first three customers are occasional, small buyers, and the last three are frequent, big spenders. Notice that fit() received only customers, with no y.

Other unsupervised tasks include dimensionality reduction (shrinking many features into a few) and anomaly detection (finding unusual items, like a strange bank transaction).

Reinforcement Learning

In reinforcement learning, an agent takes actions in an environment and gets a reward or a penalty. Over many tries, it learns which actions give the most reward.

• A game-playing program gets points for winning
• A robot gets a reward for moving forward without falling
• Chat models are partly tuned with feedback from people, rating which answers are better

Reinforcement learning needs special libraries and a lot of trial and error, so scikit-learn does not cover it. This tutorial focuses on supervised and unsupervised learning, which solve most everyday business problems.

Which Type Should I Use?

• You have past data with the answer you want to predict: supervised
• The answer is a category: classification
• The answer is a number: regression
• You have data but no answers and want to find groups: unsupervised
• A system must learn by acting and getting feedback: reinforcement

Exercise