next up previous contents
Next: Gribskov average score Up: Results for training and Previous: Zero-offset

Pseudocounts

The pseudocounts were optimized for |s|=0 through |s|=3, both separately, and minimizing the average entropy for all four sample sizes combined. The pseudocounts themselves are presented in Table 3.1, and the excess entropy for each is given in Table 6.2. The pseudocount regularizers do much better than the zero-offset regularizers for |s|=0 and |s|=1, but already by |s|=5, the difference is only 0.025 bits per column.

   table379
Table 6.2: Excess entropy for pseudocount regularizers applied to the full blocks database.



Rey Rivera
Thu Aug 1 17:59:45 PDT 1996