does 0.9.1.2 need new training db?

[Matthias Andree]

Greg Louis schrieb am Samstag, den 07. Dezember 2002:

> I haven't had time to check extensively since that discovery was made,
> but running -R on one message from both my 0.8.0 and the 0.9.1.2 has
> shown that they do _not_ give identical results.  The differences,
> however, appear in and after the third significant digit of the f(w)
> value; for example, the word "shipping" has an f(w) of 0.772122 in one
> case and 0.774 and change in the other.

With identical .db files? If so, we might have a first significant
precedence of rounding error propagation...

> My version reports S=0.000009,
> 0.9.1.2 says 0.000013.

...as the relative error is near 50% here.