This project explores the problem of rank aggregation—combining multiple ranked lists into a single consensus ranking—under conditions of partial or missing data. It implements and evaluates classical algorithms such as:

• Naive Aggregation
• Fagin’s Algorithm
• Threshold Algorithm (TA)
• No Random Access (NRA)

To address inefficiencies in NRA, particularly when the number of desired results ( k ) is large, we introduce a variant called NRA w/ Imputer, which uses imputation to estimate missing scores and enable earlier stopping.

To get started, ensure you're using Python 3.8, and follow these steps:

1. Create a virtual environment (recommended):

python3.8 -m venv venv
source venv/bin/activate   # On Windows: venv\Scripts\activate

2. Install dependencies:

pip install -r requirements.txt

This will install all necessary libraries including pandas, scikit-learn, matplotlib, numpy, and more as needed for running aggregation experiments and visualizations.

You can run any experiment using a YAML config file or by explicitly passing arguments on the command line.

python main.py --config configs\config_toy_fagin.yaml

Example for NRA w/impute algorithm:

python main.py --config configs\exp_all_algs\config_nra_impute.yaml

This will train the imputers and will save them in the .cache directory.

python main.py \
  --agg_function_name avg \
  --aggregator_name fagin \
  --dataset_name toy \
  --imputer_name basic \
  --p_erase 0.5
  --k 3 \
  --seed 42 \
  --output_path outputs/toy_fagin_p_0_5

💡 For a full list of supported arguments and defaults, refer to config.py, which defines all experiment options using a structured dataclass.

After running an experiment, results will be saved inside the folder specified by output_path. You will find:

• results.csv: Contains the number of sorted accesses and random accesses for each algorithm and value of ( k ).
• metrics.csv: Includes evaluation metrics such as set accuracy (correct top-$k$ elements) and exact match rate (correct order among top-$k$).

To generate infographics about distribution, you can run:

python data/visualize.py

The dataset is taken from https://www.kaggle.com/datasets/gregorut/videogamesales/data.

The visualizations are saved in the outputs/data directory.

If you find this work useful, please consider citing:

@misc{yehezkel2025mlrankagg,
  title={Rank Aggregation with ML-Based Imputation},
  author={Shai Yehekzel},
  year={2025},
  howpublished={\url{https://github.com/kariander1/RankAggML}},
  note={Accessed: 2025-04-21}
}