Table 4 Improvements of cBW + model selection, cBW alon vs Scheffer et al., with fully connected initial transition matrix for synthetic data with posterior-Viterbi algorithm
From: A new algorithm to train hidden Markov models for biological sequences with partial labels
State #/training sample | Average improvements of cBW + model selection in setting 1/2 | Average p value of cBW + model selection in setting 1/2 | Average improvements of cBW alone in setting 1/2 |
|---|---|---|---|
3/1600 | 7.08/8.02% | 3.6E−02/1.9E−04 | 7.49/8.35% |
3/2000 | 7.93/8.57% | 9.5E−03/3.9E−05 | 8.32/8.88% |
3/2400 | 8.21/8.85% | 1.7E−03/2.5E−05 | 8.55/9.03% |
3/2800 | 8.43/9.15% | 1.8E−03/1.1E−06 | 8.82/9.36% |
5/1600 | 9.13/10.62% | 2.4E−02/3.2E−03 | 9.08/10.58% |
5/2000 | 10.32/11.68% | 1.4E−02/6.2E−05 | 10.38/11.76% |
5/2400 | 11.26/12.74% | 1.1E−02/2.0E−06 | 11.29/12.84% |
5/2800 | 12.48/13.76% | 9.8E−03/2.1E−08 | 12.56/13.86% |
7/1600 | 8.40/10.08% | 6.0E−02/2.5E−02 | 7.77/9.50% |
7/2000 | 10.22/11.96% | 3.2E−02/1.3E−04 | 9.82/11.65% |
7/2400 | 11.10/12.68% | 8.1E−03/1.5E−05 | 10.60/12.31% |
7/2800 | 12.18/13.89% | 1.6E−04/8.4E−08 | 11.96/13.77% |