Darren Cook wrote:
Hi Darren,Just contemplating the end game a little. Seems like their are several open problems.Maybe my last ACG paper is of interest: http://www.cs.unimaas.nl/~vanderwerf/pubdown/learning_to_score.pdf
This is an interesting paper. It was nice to see comparisons of training methods in a context I'm familiar with :-).
Are you making your training dataset (with human scoring mistakes corrected) available? It sounds like you had to put in a lot of effort to create it, and I'm sure many of us would appreciate being saved that same effort. Also making it public domain means more eyeballs might discover any remaining scoring mistakes you missed.
Darren