Ranky

Compute rankings in Python.

Get started

pip install ranky

import ranky as rk

Read the documentation.

Main functions

The main functionalities include scoring metrics (e.g. accuracy, roc auc), rank metrics (e.g. Kendall Tau, Spearman correlation), ranking systems (e.g. Majority judgement, Kemeny-Young method) and some measurements (e.g. Kendall's W coefficient of concordance).

Most functions take as input 2-dimensional numpy.array or pandas.DataFrame objects. DataFrames are the best way to keep track of the names of each data point.

Let's consider the following preference matrix:

Each row is a candidate and each column is a judge. Here are the results of rk.average_rank(matrix), computing the mean rank of each candidate:

We can see that candidate2 has the best average rank among the four judges.

Let's display it using rk.show(rk.average_rank(matrix)):

In code:

import pandas as pd
import ranky as rk

m = pd.read_csv('data/matrix.csv', index_col=0)  # candidates x judges
rk.average_rank(m)         # Borda count, lower is better
rk.pairwise(m)             # Copeland's method, higher is better
rk.kendall_w(m, axis=1)    # agreement between judges
rk.show(rk.average_rank(m))

Conventions

• A preference matrix has one row per candidate and one column per judge. A judge can be a voter, a dataset, a metric, etc.
• By default, higher scores are better. Many functions have a reverse argument to change this.
• Ranking systems take an axis argument, which is the axis of the judges (axis=1 by default). Some lower-level functions (rk.rank, rk.consensus, rk.evolution_strategy, rk.brute_force) take the axis of the candidates instead. Check the docstrings when in doubt.
• With a pandas.DataFrame as input, the results keep the names of the candidates.

Ranking systems

The rank aggregation methods available include:

• Random Dictator: rk.dictator(m)
• Score Voting (mean): rk.score(m)
• Borda Count (average rank): rk.average_rank(m) or rk.borda(m)
• Majority Judgement (median): rk.majority(m)
• Uninominal Voting (multi-turn instant-runoff): rk.uninominal(m, turns=2)
• Pairwise methods. Copeland's method: rk.pairwise(m) or rk.copeland(m), Success rate: rk.pairwise(m, wins=rk.success_rate) and more. You can specify your own "wins" function or select one from the rk.duel module.
• Optimal rank aggregation using any rank metric: rk.center(m), rk.center(m, method='kendalltau'). Solver used [1].
• (Kemeny-Young method is optimal rank aggregation using Kendall's tau as metric: rk.kemeny_young(m).)
• (Optimal rank aggregation using Spearman correlation as metric is equivalent to Borda count.)

Metrics

You can use rk.any_metric(a, b, method) to call a metric from any of the three categories below.

• Scoring metrics: rk.metric(y_true, y_pred, method='accuracy'). Methods include: ['accuracy', 'balanced_accuracy', 'precision', 'average_precision', 'f1_score', 'mxe', 'recall', 'jaccard', 'roc_auc', 'mse', 'rmse', 'mae', 'sar']

• Rank correlation coefficients: rk.corr(r1, r2, method='spearman'). Methods include: ['kendalltau', 'spearman', 'pearson']

• Rank distances: rk.dist(r1, r2, method='levenshtein'). Methods include: ['hamming', 'levenshtein', 'kendall', 'euclidean', 'winner', 'winner_mistake', 'winner_distance', 'symmetrical_winner_distance']

To add: general edit distances, kemeny distance, regression metrics...

Visualizations

• Use rk.show to visualize preference matrix (2D) or ranking ballots (1D).

>>> rk.show(m)

>>> rk.show(m['judge1'])

• Use rk.mds to visualize (in 2D or 3D) the points in a given metric space. See rk.scatterplot documentation for display arguments.

>>> rk.mds(m, method='euclidean')

>>> rk.mds(m, method='spearman', axis=1)

• You can use rk.tsne similarly to rk.mds.

• Use rk.critical_difference to plot a critical difference diagram, comparing candidates' performance and grouping them by statistical equivalence. Such diagrams can be seen in [2, 3].

>>> rk.critical_difference(m, comparison_func=rk.bayes_wins)

• Show Condorcet graphs using rk.show_graph(graph), based on [4]. The graph can be computed with rk.pairwise(m, return_graph=True).

Other

• Rank, rk.rank, convert a 1D score ballot into a ranking.
• Bootstrap, rk.bootstrap, sample a given axis.
• Consensus, rk.consensus, check if rankings exactly agree.
• Concordance, rk.concordance, mean rank correlation between all judges of a preference matrix.
• Centrality, rk.centrality, mean rank correlation (or distance) between a ranking and a preference matrix.
• Kendall's W, rk.kendall_w, coefficient of concordance.
• Judge generators, rk.SwapGenerator and rk.GaussianGenerator, sample noisy judges from a reference ranking.
• Utility: rk.read_codalab_csv to parse a CSV generated by Codalab representing a leaderboard into a pandas.DataFrame.

Development

See CONTRIBUTING.md to run the tests, build the documentation and release a new version. The changes between versions are listed in CHANGELOG.md.

References

Please cite ranky in your publications if this is useful for your research. Here is an example BibTeX entry:

@misc{pavao2020ranky,
  title={ranky},
  author={Adrien Pavao},
  year={2020},
  howpublished={\url{https://github.com/didayolo/ranky}},
}

[1] Storn R. and Price K., Differential Evolution - a Simple and Efficient Heuristic for Global Optimization over Continuous Spaces, Journal of Global Optimization, 1997, 11, 341 - 359.

[2] Janez Demsar, Statistical Comparisons of Classifiers over Multiple Data Sets, 7(Jan):1--30, 2006.

[3] H. Ismail Fawaz, G. Forestier, J. Weber, L. Idoumghar, P. Muller, Deep learning for time series classification: a review, Data Mining and Knowledge Discovery, 2018.

[4] Aric A. Hagberg, Daniel A. Schult and Pieter J. Swart, “Exploring network structure, dynamics, and function using NetworkX”, in Proceedings of the 7th Python in Science Conference (SciPy2008), Gäel Varoquaux, Travis Vaught, and Jarrod Millman (Eds), (Pasadena, CA USA), pp. 11–15, Aug 2008.

License

Copyright (c) 2020-2021, Adrien PAVAO. This software is released under the Apache License 2.0 (the "License"); you may not use the software except in compliance with the License.

The text of the Apache License 2.0 can be found online at: http://www.opensource.org/licenses/apache2.0.php