FlexMF Logistic on ML100K

This page analyzes the hyperparameter tuning results for the FlexMF scorer in implicit-feedback mode with logistic loss (Logistic Matrix Factorization).

Parameter Search Space

Parameter	Type	Distribution	Values
embedding_size	Integer	LogUniform	4 ≤ \(x\) ≤ 512
regularization	Float	LogUniform	1e-05 ≤ \(x\) ≤ 1
learning_rate	Float	LogUniform	1e-05 ≤ \(x\) ≤ 0.1
reg_method	Categorical	Uniform	L2, AdamW
negative_count	Integer	Uniform	1 ≤ \(x\) ≤ 5
positive_weight	Float	Uniform	1 ≤ \(x\) ≤ 10

Final Result

Searching selected the following configuration:

{
    'embedding_size': 31,
    'regularization': 0.39291404142015096,
    'learning_rate': 0.09343906641585487,
    'reg_method': 'AdamW',
    'negative_count': 4,
    'positive_weight': 2.096370093026729,
    'epochs': 22
}

With these metrics:

{
    'RBP': 0.12731247414883756,
    'NDCG': 0.3980651998294027,
    'RecipRank': 0.3880723447117054,
    'TrainTask': '7feaf81d-97d6-4610-ba8c-c48acc0ce47e',
    'TrainTime': 5.025425890999031,
    'TrainCPU': 4.984771,
    'timestamp': 1743028571,
    'checkpoint_dir_name': None,
    'done': True,
    'training_iteration': 22,
    'trial_id': '6099e_00061',
    'date': '2025-03-26_18-36-11',
    'time_this_iter_s': 0.21425771713256836,
    'time_total_s': 5.145690441131592,
    'pid': 159198,
    'hostname': 'CCI-ws21',
    'node_ip': '10.248.127.152',
    'config': {
        'embedding_size': 31,
        'regularization': 0.39291404142015096,
        'learning_rate': 0.09343906641585487,
        'reg_method': 'AdamW',
        'negative_count': 4,
        'positive_weight': 2.096370093026729,
        'epochs': 22
    },
    'time_since_restore': 5.145690441131592,
    'iterations_since_restore': 22,
    'experiment_tag': 
'61_embedding_size=31,learning_rate=0.0934,negative_count=4,positive_weight=2.0964,reg_method=AdamW,regularization=
0.3929'
}