What are these inputs in EvaluateCorrect.m ?
train_users, test_users
Are they the cf-train-1/10-users.dat, cf-test-1/10-users.dat stored in data folder?
if yes,repmat(num_total, 1, M) may become zeros, as some rows of test users are zero. There are some users with only 10 items and all of these were selected for training.
After removing NAN rows from (cumsum(num_hit, 2)./repmat(num_total, 1, M);) my evaluation (mean(cumsum(num_hit, 2)./repmat(num_total, 1, M),1) becomes 17%. The interesting point is that the result for sparse dataset is 8. both of them are1/4 reported results in the papers. I do not know why.
note: my training procedure cannot decrease neither "loss" nor "gen_loss". it remains unchanged from the first epoch of test_cvae to epoch no 500.
What are these inputs in EvaluateCorrect.m ? train_users, test_users Are they the cf-train-1/10-users.dat, cf-test-1/10-users.dat stored in data folder? if yes,repmat(num_total, 1, M) may become zeros, as some rows of test users are zero. There are some users with only 10 items and all of these were selected for training. After removing NAN rows from (cumsum(num_hit, 2)./repmat(num_total, 1, M);) my evaluation (mean(cumsum(num_hit, 2)./repmat(num_total, 1, M),1) becomes 17%. The interesting point is that the result for sparse dataset is 8. both of them are1/4 reported results in the papers. I do not know why. note: my training procedure cannot decrease neither "loss" nor "gen_loss". it remains unchanged from the first epoch of test_cvae to epoch no 500.