Find the groups yourself

The idea

Clustering is when a machine puts similar items together without being given labels, which is how it differs from classification.

Sort them with no labels

A coach has season numbers for nine players: runs scored and wickets taken. Nobody has said who is a batter, a bowler or an all-rounder. How could you sort them?

• Put players with similar runs and wickets together
• Someone must first tell me each player's type
• It cannot be done with only numbers

Let the machine group them

Each dot is a player from a school team. Across is runs scored and up is wickets taken. Nobody has labelled them.

Press Use suggested start to place three centres. Do not tap the plane yet. Then press Step again and again, and watch the dots change colour.

Groups with no labels

Look at where each group sits. Bottom right has high runs and few wickets: batters. Top left: bowlers. The middle: all-rounders. The machine was never told those words.

It only saw runs and wickets. Each round, every player joined the nearest centre, then each centre moved to the middle of its players. When nothing changed, it stopped. We named the groups afterwards.

Clustering means a machine puts similar items together without being given labels, unlike classification.

Learning without labels is called unsupervised learning.

Notes

Clustering means a machine puts similar items together without being given labels, unlike classification.

A person must name and check the groups. Badly placed starting centres can give different groups, so real systems try several starts.

Check yourself

A kirana shop knows each customer's visits per month and money spent, but it has no labels. It wants groups of similar customers. Which job is this?

A music app groups songs by speed and loudness and finds four groups. Who gives each group its name?

A school app groups students by marks and homework time. Group 3 has the lowest marks. What is the fair way to use this?

• Classification. Classification needs groups that were labelled in advance. This shop has no labels at all.
• Regression. Regression predicts a number. The shop wants groups, not a number.
• Clustering — correct. Yes. With no labels, the machine can only put customers who look alike together. People then decide what each group means.
• The machine, because it learned the names from the songs. Nobody gave it any names. It only compared numbers.
• A person, who listens to each group and picks a name — correct. Yes. The machine shows which songs are alike, and a person decides what that means.
• The music company, because it owns all the songs. Owning the songs does not tell what the groups mean. The groups come from speed and loudness numbers, so someone must listen and judge.
• As a hint about who may need help, then talk with each student — correct. Yes. A group is a pattern in a few numbers. It cannot tell what a student is able to do.
• Tell the students in Group 3 that they are slow learners. Marks and homework time do not show how well someone can learn. A label like that can hurt.
• Show every student's group on the class board. Marks are private. Showing groups to everyone could embarrass students without their say.