PASCAL Agnostic Learning
vs.
Prior Knowledge
IJCNN07

The challenge is now over. But it remains open for post-challenge submissions!


IMPORTANT: Entries made since February 1st 2007 might be using validation data, now available for training.

cross-indexing-1

Submitted by Juha Reunanen

The not-too-promising performance guesses:

- ADA: 0.243445
- GINA: 0.0698001
- HIVA: 0.320429
- NOVA: 0.0652942
- SYLVA: 0.0173416

Models were selected from amongst CLOP models. The problem however is (probably) that the most powerful models are not included in the search space. And maybe also that cross-indexing was designed specifically for feature selection -- the adaptation to generic model selection clearly is not ready yet.

Dataset Balanced Error Area Under Curve  
Train Valid Test Train Valid Test
ada 0.1773 0.2161 0.1964 0.8247 0.8104 0.8045 agnostic
gina 0 0.0285 0.0436 1 0.9706 0.9571 agnostic
hiva 0.1335 0.405 0.3333 0.8575 0.5701 0.6691 agnostic
nova 0.0004 0.054 0.0534 0.9993 0.937 0.9469 agnostic
sylva 0.0048 0.0041 0.011 0.9947 0.9957 0.989 agnostic
Overall 0.0632 0.1415 0.1276 0.9352 0.8568 0.8733 agnostic

This entry is a Model Selection Game entry.