Table 4 Forming a consensus labelling. Forming a consensus labelling from k paths. Results for Viterbi decoding and 1-best decoding (as used in Phobius) are provided for comparison. (n = 247).
From: Decoding HMMs using the k best paths: algorithms and applications
| Â | Ï„= 0 | Ï„= 1 | Ï„= 2 | Ï„= 3 | Ï„= 4 | Ï„= 5 | Phobius |
|---|---|---|---|---|---|---|---|
k = 10 | 3 | 12 | 23 | 27 | 47 | 69 | 139 |
k = 100 | 4 | 14 | 23 | 32 | 51 | 70 | 141 |
k = 1000 | 3 | 15 | 24 | 35 | 57 | 77 | 145 |
Viterbi (k = 1) | 2 | 11 | 19 | 27 | 46 | 59 | 137 |
1-best | 4 | 18 | 26 | 38 | 62 | 79 | 166 |