Set | Model | POP | IPA | >Acc |
---|
Bali
|
g2
| 80.3% | 80.0% | 0.4% |
Bali
|
g4
| 80.6% | 79.6% | 1.3% |
TermGaps
|
g2
| 80.2% | 79.7% | 0.6% |
TermGaps
|
g4
| 84.0% | 83.8% | 0.2% |
- Training set accuracy (Q) for the g2 and g4 models on Bali and TermGaps using BLOSUM62 1/3-bit units. in The numbers in the POP and IPA columns are the Q values achieved by the gap para-meters reported by each method. In all cases, POP achieves higher accuracy, the >Acc column shows the amount by which the POP accuracy is better than IPA.