Hi,
I am trying out different algos with Movielens 100k (ua.base and ua.test).
But best rmse I could find is biassgd
./toolkits/collaborative_filtering/biassgd --training=movielense.train --validation=movielense.test --biassgd_lambda=0.25 --biassgd_gamma=0.10 --minval=1 --maxval=5 --max_iter=100 --quiet=1
Iteration: 99 Training RMSE: 0.906892 Validation RMSE: 0.955748
Can you post best some best tunings ... as I have seen some here http://mymedialite.net/examples/datasets.html but could not translate them directly here.
Also, how to control number of latent features (hidden layer size in case of RBM) ?