This page answers short questions about the ranked chooser in this book. Each answer is a few lines long. Each one links to the section that explains it in full.
Contents#
- Can the trees work when the camera sees only part of the table?
- Why do the trees lose to a plain rule here?
- Is it faster, since a separated table takes it less time?
- Would charging for push length fix the order?
- Why was the corrected jaw input not kept?
1. Can the trees work when the camera sees only part of the table?#
Yes. Their inputs are relations between glasses, not places on the table. And the arm looks round the whole table before it chooses. Only the rack input lost its meaning. The trees can still read the table gives the details.
The next question is why they still lose.
2. Why do the trees lose to a plain rule here?#
They learned on tables where every glass with room had been lifted away. Here those glasses stay and get in the way. The trees lean on the glass's shape, not on the room around it. Why the trees choose worse here gives the numbers.
The next question is about one number that looks like a win.
3. Is it faster, since a separated table takes it less time?#
No. That average counts only won tables. On the 6 tables both solutions separated, this one took 29% more arm time. Where the three tables went explains it.
The next question is about the first repair that was tried.
4. Would charging for push length fix the order?#
No. Four prices were tried on 40 tuning tables. The number of tables separated stayed at 14 or 15 at every price. Two repairs that were measured and not kept has the table.
The last question is about the second repair.
5. Why was the corrected jaw input not kept?#
It helped in perfect play, but it lost under the clock: 12 tables against 14. Choices are made on a clocked queue, as where choices are made explains. So it was dropped. The labels still come from the old jaw. Two repairs that were measured and not kept gives both results.
The next chapter is the first solution that learns from this solution's answers, and it asks what copying them buys.
← What is different · Imitation from demonstrations — what is different →