← Back to course
AI Explorers 6-8 / Week 05 / Explorer Quest
6/6
Week 05 · Rules or Learning?

Explorer Quest

Explorer Quest: Find the family cut-off
// Hand-written rules versus patterns found in data
⏱ about 20 min

Explorer Quest: Find the Family Cut-off

Raven has a puzzle for the whole crew. She lines up a row of paper cups, each with some water in it.

"Some are labeled full enough to plant seeds, and some are labeled too low," she says.

Rocket grabs a ruler. "I will just guess: anything over three centimeters is enough."

"Do not guess," Nova says. "Measure every labeled cup and count which cut-off matches the most labels."

Raven hands Rocket a notebook. "Let the data decide, then test it on a new cup."

  1. Hand-written rules are chosen by people. They are clear but can break on tricky cases.
  2. Machine learning is a kind of statistical inference that finds patterns in data.
  3. Learning needs tremendous amounts of training data, usually supplied by people.
  4. Compare methods fairly by testing them on the same examples and counting right answers.
◇ FIND THE FAMILY CUT-OFF
Pick a measurement you can take at home, like the height of ten books or the length of ten socks.
One person secretly labels each item "big" or "small" using their own sense of it.
Everyone else measures every item with a ruler and writes the numbers in a table.
Test three possible cut-offs. Count how many labels each cut-off matches.
Choose the cut-off the data supports. Then test it on two new items.
For a grown-up
Use safe, everyday objects and a ruler. Nothing sharp or breakable.
Ask: was the best cut-off the one you would have guessed?
Ask: what would happen if the labeler changed their mind about one item?
QUEST CHECK
  • Read the question.
  • Tap your answer.
Cut-off X matches 7 of 10 labels. Cut-off Y matches 9 of 10. Which does the data support?
Who supplied the labels in your family game?
Finding patterns in data by counting and comparing is called statistical ____.
WHY THIS EXERCISEMachine learning is a kind of statistical inference.
← Friday