WCCI Performance Prediction Challenge

The challenge is over, but a new challenge is on-going using the same datasets, check it out!

BNN-DRIPC

Submitted by Radford Neal

Bayesian neural network, with two final hidden layers
that look at a earlier linear hidden layer that reduces
dimensionality from most or all features, along with
some principal components. Some special stuff is also
done for HIVA and NOVA.

Dataset Balanced Error Test guess Guess error Test score Area Under Curve
Train Valid Test Train Valid Test
ada 0.1458 0.2027 0.1755 0.165 0.0105 0.1859 0.9321 0.8993 0.9098
gina 0 0.0351 0.0452 0.0646 0.0194 0.0646 1 0.9874 0.9903
hiva 0.1178 0.3452 0.291 0.2691 0.0219 0.312 0.9432 0.6782 0.7646
nova 0 0.076 0.0504 0.0693 0.0189 0.0693 1 0.9824 0.9894
sylva 0.0044 0.0037 0.007 0.0074 0.0004 0.0072 0.9998 0.9994 0.9992
Overall 0.0536 0.1325 0.1138 0.1151 0.0142 0.1278 0.975 0.9093 0.9306

This entry is a complete valid challenge entry.