This page answers the questions a reader most often asks about solution 6 in this problem. Each answer is short. Each one links to the section that explains it in full.
Contents#
- Why draw each batch evenly, instead of changing the demonstrations?
- What does "work, but not care" mean in numbers?
- The paths got closer to the teacher's, so why did the score not move?
- Why was the fit with three times the steps not kept?
- How is this fit the mirror image of the pushing problem's fit?
1. Why draw each batch evenly, instead of changing the demonstrations?#
The demonstrations record what the teacher really did. Four answers in five are looks, and that is a fact about this problem. So the set is kept as it is. Only the batches are drawn half looks and half pushes. Two changes, and neither is a new model describes it.
2. What does "work, but not care" mean in numbers?#
On the same twenty tables, the trained model freed two glasses and the untrained one freed none. The trained model also broke two glasses, and the untrained one broke none. Neither separated a table. The pair: work, but not care gives the table.
3. The paths got closer to the teacher's, so why did the score not move?#
The training taught the size and speed of a push. It did not teach where to put the hand down. The teacher mostly looks, so a close copy mostly looks too. The paths took the teacher's shape shows the measurements.
4. Why was the fit with three times the steps not kept?#
It brought the paths closer still. But it broke as many glasses on the twenty tables, and it scored lower on the held-out queue. It also cost three times the machine time. Three times the steps moved the paths, not the table has the numbers.
5. How is this fit the mirror image of the pushing problem's fit?#
There, every demonstration was a push, and the fit learned the height of a push but not its length. Here, most are looks. With an even mixture, the fit learned the length and speed but not the height. The mirror image of the pushing problem compares them.
← What is different · The results →