Skip to content
Exit presentation

How AI Learns I: From Rules to Data

Keep some crates back

The evening pile has 24 priced crates. Which ones should the program learn from?

Loading lab…

Twenty-four crates, one evening pile.
Step 1 of 4
  1. Twenty-four crates, one evening pile.
  2. Split: every third crate goes to the kept-back pile.
  3. Learn pile: 16 crates. Kept-back pile: 8 crates.
  4. The program never sees the kept-back pile.

The rule

Learn from most crates. Keep a few back, unseen.

One split makes the learn set and the kept-back set. A kept-back crate is one the program has never seen. Crates it never saw test it fairly.

  • Training data — the learn set. The crates the program is shown.
  • Test data — the kept-back set. The crates it never sees.

The score is how many kept-back crates the program guessed right. The lightly marked half is every crate with 7 or fewer marked.

Three splits of the same evening pile
SplitLearnKept backScore
Every third crate1684 of 8
First eight crates1682 of 8
Lightly marked half11132 of 13

The program is the same in every row. Only the split moved. So say how you split, and anyone can repeat the score.

The first eight crates are all full price. The program learns from the other 16. Of these, 12 are cut price. So the program guesses cut too often. It gets only 2 of the 8 right.

Jump to lesson

This track

  1. 0 Start here
  2. 1 A rule a person wrote, and a pile of examples
  3. 2 Keep some crates back
  4. 3 Right out of how many
  5. 4 Split it differently
  6. 5 Learning the examples instead of the pattern
  7. 6 Ask more than one
  8. 7 The order you do it in
  9. 8 Write the rule yourself
  10. 9 The big picture
  11. 10 Track quiz

Other tracks