Bias Predictor

This page analyzes the behavior and hyperparameter search for the Bias scoring model on ML10M.

Parameter Search Space

Parameter Type Distribution Values Selected
damping.user Float LogUniform 0.1 ≤ \(x\) ≤ 100 41.7
damping.item Float LogUniform 0.1 ≤ \(x\) ≤ 100 6.47

Final Result

Searching selected the following configuration:

{'damping': {'user': 41.676997394336546, 'item': 6.469202016377055}}

With these metrics:

{
    'RBP': 0.09411545753870114,
    'LogRBP': 1.4310069910393786,
    'NDCG': 0.39269981890244177,
    'RecipRank': 0.26514221927427595,
    'RMSE': 0.7833195825393295,
    'TrainTask': '8f74f689-4165-4f5d-ae62-c10b4c1b61d6',
    'TrainTime': 1.81669005099684,
    'TrainCPU': 1.8213460000000001,
    'TestTask': '3ad4bc7a-ff9f-4acb-a0d0-012030f2e730',
    'TestTime': 3.9537269709981047,
    'TestCPU': 3.9512750000000003,
    'timestamp': 1746383978,
    'checkpoint_dir_name': None,
    'done': True,
    'training_iteration': 1,
    'trial_id': '88b07c8e',
    'date': '2025-05-04_14-39-38',
    'time_this_iter_s': 8.263916492462158,
    'time_total_s': 8.263916492462158,
    'pid': 2372609,
    'hostname': 'CCI-ws21',
    'node_ip': '10.248.127.152',
    'config': {'damping': {'user': 41.676997394336546, 'item': 6.469202016377055}},
    'time_since_restore': 8.263916492462158,
    'iterations_since_restore': 1,
    'experiment_tag': '77_item=6.4692,user=41.6770'
}

Metric Response

How does RMSE change with each setting independently?

Best Configurations

Since this is an explicit-feedback rating prediction model, our primary search criteria is RMSE. The configuration with the best RMSE is:

config.damping.user config.damping.item RBP RMSE
Method
Random 41.676997 6.469202 0.094115 0.78332

If we instead searched for RBP, we would select:

config.damping.user config.damping.item RBP RMSE
Method
Random 5.571412 99.639145 0.105656 0.790722

Search Geometry

What is the geometry of the search space?