This experiment is mentioned in the intuitive explanation of State Grouping in the implementation section.
To compute the individual responsibilities, run:
./target/release/bw-responsibility --prism-model experiments/train_grouped/train_grouped.prism -c "experiments/train_grouped/train_grouped.ce" -b "sbar"You should get the following responsibilities:
| State | Responsibility |
|---|---|
| (s=2) | 0.52380952 |
| (s=1) | 0.35714286 |
| (s=5) | 0.02380952 |
| (s=6) | 0.02380952 |
| (s=7) | 0.02380952 |
| (s=8) | 0.02380952 |
| (s=9) | 0.02380952 |
To compute grouped responsibilites, run:
./target/release/bw-responsibility –prism-model experiments/train_grouped/train_grouped.prism -c “experiments/train_grouped/train_grouped.ce” -b “sbar” -g
You should get the following responsibilities:
| State | Responsibility |
|---|---|
| (s2) | 0.66666667 |
| (s1) | 0.16666667 |
| (t) | 0.16666667 |