WCCI Performance Prediction Challenge

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

BNN-DRb

Submitted by Radford Neal

Bayesian neural network with three hidden layers. The first layer is
linear, and reduces dimensionality to ten, the following layers allow
non-linear mappings from this ten dimensional space. Some special
stuff is done for NOVA. This version is the same as BNN-DR except the
MCMC runs are longer.

Dataset Balanced Error Test guess Guess error Test score Area Under Curve
Train Valid Test Train Valid Test
ada 0.1442 0.1898 0.1753 0.1652 0.0101 0.1853 0.9312 0.9027 0.9099
gina 0 0.0572 0.047 0.0597 0.0127 0.0597 1 0.9841 0.9883
hiva 0.0345 0.3809 0.2915 0.2684 0.0232 0.314 0.9929 0.7405 0.7582
nova 0 0.058 0.0511 0.1059 0.0548 0.1059 1 0.9808 0.9882
sylva 0.004 0.0037 0.0066 0.0072 0.0006 0.0071 0.9999 0.9992 0.9991
Overall 0.0365 0.1379 0.1143 0.1212 0.0203 0.1344 0.9848 0.9215 0.9287

This entry is a complete valid challenge entry.