From: Adam S. <crm...@ad...> - 2005-02-14 16:43:44
|
Fidelis Assis (fi...@po...) wrote: > I see about the same accuracy with toer.crm as you and that is very good > for a single pass over about 5000 messages - 99.6% => 2 errors/500. Hmm, when you put it like that ... 2 errors ain't bad :-) > See, you could have had only 2 results better than that: 1 error/500 => > 99.8% and 0 error => 100%. I guess that even with a single error one > would be disappointed when comparing to 99.984%. And 0 error isn't a > real result, of course :). Good point :-) OK, I will try multiple shuffles and bigger corpus. |