Learning from labelled examples

Your goal

Explain how examples with answers help a model sort a new leaf.

To be able to explain how labelled examples help a model classify new items.

Before you start

A leaf has a toothed edge and belongs to Type B. Which is the feature?

Choices: Toothed edge / Type B

Words for this lesson

Feature

A detail measured or noticed in an example.

Label

The known answer attached to a training example.

Training

Using examples to set a model's numerical values.

The idea

People provide examples with known answers. This is called supervised learning. A feature is a detail such as the number of teeth on a leaf edge. A label is the leaf type. During training, a model uses the examples to set numerical values. It then predicts labels for different examples. We compare its predictions with the known answers. A test item must not be used to train this version.

A completed example

A practice sorter sees two labelled objects: a 2 cm stick labelled Short and an 8 cm stick labelled Long. Their lengths are features. Short and Long are labels. Testing a different stick checks whether the pattern helps beyond the practice examples. This example is separate from the leaf task.

Check your understanding

Which is a label?

1. Type A
2. Four centimetres
3. Six edge teeth

Which examples test this trained version?

1. Different leaves kept out of training
2. Only A1 again
3. The teacher's opinion

Does adding more examples guarantee no mistakes?

1. No, their quality and variety matter
2. Yes, always
3. Yes, if labels are hidden

Task 1 · Make a prediction

Prediction for T1

Options: Type A / Type B / Needs review

Write your prediction, or state that you skipped it.

Try it using supplied records

1. Read leaf A1. Select its feature and label. Read the other three labelled examples.
2. Predict whether new leaf T1 is Type A or Type B. Press Train model. The site calculates one centre for each type.
3. Press Test new leaves. Compare the four predictions with their known answers.
4. Change A2 to the wrong label Type B. Retrain and retest the same four leaves. Record which prediction changed. Restore A2 afterwards.

Real small algorithm · nearest-centre training

Complete data and source records

Expected results · supplied

These are supplied reference results. They are separate from your own explanation. No code needs to run to read them.

1. Implement a nearest-centre classifier. Training computes the arithmetic mean of edge teeth and length separately for each label. Predict the label with the smallest squared Euclidean distance. Keep these units and features fixed for this toy dataset. Resolve equal distances as Needs review. Show distances only in More detail.

2. Correct labels give centres A=(0,3), B=(6,3), predictions A,B,A,B and 4/4 correct. With A2 relabelled B, centres A=(0,2), B=(4,10/3); predictions A,B,B,B and 3/4 correct. Retraining replaces centres; testing never updates them. Disable training if a class has no examples.

3. States: examples -> prediction recorded -> trained -> tested -> changed label -> retrained -> compared. A train action clears old test results and marks them stale until retested. Store both run records with the label state.

Help with this step

Task 2 · Use the evidence

1. Complete: The labels in our training set are __. One measured feature is __.
2. Use the two test records to explain how changing a label affected T3.
3. Explain why a new test leaf is more useful than repeating only a training leaf. Give one sensible check before using the classifier on real leaves.

My explanation

Write in your school assignment. Keep these answers in your own words.

Your explanation

• Training labels
• A measured feature
• What happened to T3 and why
• Why a new test helps and my next check

Explain a fresh example

A model predicts Type B for a new leaf. Explain how labelled examples helped it make that prediction and name one check you would make.

My confidence

I need another explanation / I can explain with help / I can explain with my own example

My next step

Write what you will do next.

Optional depth

Would four perfect test answers prove this classifier works for all plants? Explain what variety is missing.

What this activity can show

This is genuine numerical training of a very small nearest-centre classifier, not a neural network. Its two invented features cannot identify real plant species reliably.

This is a student learning record. Submission is completed through the school assignment system.