ViLLA-MMBench Documentation

Grid Search Module (grid.py)

Functions:

• gridMetric(model, val_grp, train_fit_set, train_seen, iid_map, all_iids, USE_GPU_FOR_HPO, topN=10)
  • Evaluates model performance using specified metrics
  • Used during hyperparameter optimization
• grid(dataDict, cls, name, scenario, param_grid, *fit_args)
  • Performs grid search for hyperparameter optimization
  • Handles model fitting and evaluation
• gridSearch(config: dict, train_df: pd.DataFrame, train_set: pd.DataFrame, modalities_dict: dict)
  • Main function for model selection and training
  • Supports multiple model types: CF, VBPR, VMF, AMR, VAECF
  • Handles different modality combinations

Processes Module (processes.py)

Functions:

• generateLists(config: dict, train_df, train_set, test_df, genre_dict, final_models)
  • Generates recommendation lists for evaluation
  • Calculates various metrics:
    • Popularity Bias (PB)
    • Fairness (FA)
    • Novelty (NO)
    • Diversity (DI)
    • Calibration Bias (CB)
    • Recall (RC)
    • NDCG (ND)
    • Coverage (CV)
  • Outputs results to CSV files

Main Process Flow

The main process flow (main.py) consists of the following steps:

1. Configuration loading from config.yml
2. MovieLens data preparation and train-test splitting
3. Loading and processing of text, visual, and audio embeddings
4. Hyperparameter tuning using grid search
5. Generation of recommendation lists and metric calculation