CBuilderProject· about 90 min· in the browser

Project 1: Teach a computer to recognise patterns

Build a fruit classifier from two measurements (weight, colour value) entirely by hand: features, distance, nearest neighbour, accuracy and error analysis.

Teaches: Project: train an image classifier in the browser

Requires: Classification: how a model sorts informationPython — lists, loops and dictionaries

Theory

Every fruit is a point (weight, colour). A new fruit gets the same label as its nearest neighbour. Accuracy is measured on fruits the model has not seen — and the mistakes it makes tell you more than the number.

Sub-tasks

features — features(frukt) → [vikt, farg] (weight, colour) as numbers from a dict.
distance and nearest — avstand(a, b) (Euclidean distance) and narmaste(punkt, traning) → the label of the nearest training fruit.
accuracy — traffsakerhet(traning, test) → fraction correct.
error analysis — felanalys(traning, test) → dict {(actual, guessed): count} for the mistakes — which fruits get confused?

Passes when: accuracy >= 0.8

The starter code

runs in your browser

from data import TRANING, TEST


def features(frukt):
    # TODO: [frukt["vikt"], frukt["farg"]]
    ...


def avstand(a, b):
    # TODO: euklidiskt avstånd mellan två featurelistor
    ...


def narmaste(punkt, traning):
    # TODO: etiketten hos den träningsfrukt vars features ligger närmast punkt
    ...


def traffsakerhet(traning, test):
    # TODO: andel av test där narmaste(features(f), traning) == f["etikett"]
    ...


def felanalys(traning, test):
    # TODO: dict {(ratt, gissat): antal} bara för felen
    ...

You write the code; tests you cannot see decide whether it holds up. Create a free account to run the lab.

Try the diagnostic Create a free account

Expected results

Accuracy ≥ 0.8 on the test fruits; the error analysis shows that apples and pears are confused most often (they have similar weights).

Common mistakes

Compares colour (0–1) and weight (grams) without thinking about scale — the weight dominates the distance. Try scaling!
Counts the test fruit itself among the neighbours.