aaagh.

[trochee]

well, the system i've been working/collaborating on with two labmates and Advisor for some five months just got through its first run on the development data.

the oracle error (best possible given the experimental conditions): 0.37
the baseline system (no bells or whistles): 0.41
our snazzy system: 0.43

Right.

0.43.

Worse than the baseline.

feck. I hope that's a bug we can find and not a failure of the algorithm.

*sigh*

Comments

[tulip-tree.livejournal.com]

Oh no! My sympathies... I know the horrible feeling of seeing results like that and wondering what went wrong. I hope it's a bug and that you find it soon. Good luck!

[beckyb.livejournal.com]

Ugh. Well, at least there is a result to debug.