Track quiz
Eight lessons, one line each. Keep this table open while you answer.
| Idea | The line to keep |
|---|---|
| A rule or a pile | One rule a person wrote and one rule learned from examples. Measured on the same kept-back set, both produce a count. The counts do not tell you which crates each one got wrong. |
| Keep some back | 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. |
| Right out of how many | Walk both lists together. Add one when a pair agrees. Hand back the count. Two lists that agree on nothing give 0. |
| Split it differently | Move the split and measure again. The count and the number kept back are reported together. 3 of 4 and 30 of 40 are not the same claim. |
| Learning the examples | A nearest-example program that holds a crate finds it at distance 0. So it copies that crate's word. That is automatic, not a skill. The only count that matters covers crates it has never seen. |
| Ask more than one | Ask one, three or five nearby examples and let the majority answer. A split vote is reported as a tie. The count shows whether more examples helped. |
| The order you do it in | Collect, label, split, learn, measure — in that order. Measure on the examples it learned from, and the count flatters. Measure on the kept-back tray, and it means something. |
| Write the rule yourself | Walk the examples once and carry the nearest one so far. Hand back its word at the end. When two are exactly as near, keep the one met first. A vote of several examples reports a tie instead. This walk keeps the first. |
Two rules score 15 and 14 on the same 20 test crates. What else should you check?
A crate is kept back. What does the program know about it?
24 crates are split. Every third crate is kept back. How many does the program learn from?
Why should you say how a split was made?
The lists agree on nothing. What does agreed hand back?
Guesses: full, cut, cut. Real words: full, full, cut. How many guesses are right?
One program scores 4 of 8 on a split. On another split, 2 of 8. What moved?
What must you write beside a count of 14?
A nearest-example program scores 24 of 24 on crates it was shown. What does that tell you?
A program gets 24 of 24 on its own crates. It gets 14 of 20 on new ones. Which counts?
The three nearest examples say cut, cut, full. What is the answer?
On one pile, asking 5 scored worse than asking 1. How can that happen?
The program learns, then is measured. Only then is the pile split. What is it tested on?
The clerk never wrote the words on the crates. What happens at the measuring step?
Your walk meets an example as near as its best. What does it keep?
A new crate is at 3 marked and 2 kilos. An example is at 6 marked and 6 kilos. What is the squared distance?
That is the track: a pile, a kept-back set, one honest number.
Ask every score what it was measured on.
Next: Using AI Well & Safely I.