I competed in another Kaggle competition called Gender Prediction from handwriting. Goal of the competition was to detect gender of a writer based on a writing style.
Error metric is LogLoss.
For this competition I tried many different models, but didn't submit a result, because no model was better then benchmark.
I played with Ramp, (please use my clone with Fixes for some things, for this code). It is very nice to work with, model testing is much faster.
I also tried Spearmint for parameter optimization but got only numpy errors when I ran it.