Results
- First we tried with only first level inputs.
| T1 | T2 | T3 | T4 | T5 | T6 |
base features | 100 | 52.3 | 75.6 | 98.3 | 73.6 | 51 |
base features | 100 | 51.2 | 76.4 | 93.6 | 73.6 | |
triple features | 100 | 100 | 100 | 100 | 100 | 100 |
- Note that type 1 and 4 are linearly separable.
- Note that type 2 and 6 have no information in the individual features.
- If we use triples alone as input, we get perfect behaviour
- Neither first or triples are what we're looking for.
- We're looking for more or less perfect results, but some should
be more difficult than others.
| T1 | T2 | T3 | T4 | T5 | T6 |
Shepard et al | first | second |
joint third | joint third | joint third |
last |
- Subjects make more mistakes, are slower to learn the categories,
and are slower to answer by this precdence.