improved representation of bias

[Soren Harward]

On Sunday 14 December 2008 21:36:53 Simon Bühler wrote:
> What was really disappointing though - even setting the similarity slider
> to 100% produces a playlist, where a randomize() would produce the same
> results; imo led zeppelin isn't really similar to redman and paul van dyk -
> i feel the fingerprint algorithm is not working in a usable way.

I assume that you ran the GTACFeat tagger program on all your audio files so 
that they'd have embedded fingerprints?  If they all have null fingerprints, 
then Amarok will think they're similar.  This behavior does need to be 
changed; that's why the code is still experimental.

And I will say that acoustic similarity calculation isn't an exact science by 
any means.  The current similarity algorithm in GTACFeat is quite simple, and 
even the best similarity calculation algorithms aren't nearly as good as a 
human.

-- 
Soren Harward
stharward at gmail.com