Github

Last update: 10/2013

Relocated from sourceforge.net/projects/fantailmlkit/

Description

Fantail is a collection of machine learning algorithms for ranking prediction, multi-target regression, label ranking and metalearning related data mining tasks. The algorithms can be called from your own Java code. It is also well-suited for developing new algorithms. Fantail is a multi-target learning extension to WEKA, and is at the early development stage. New algorithms and tools will be added to the library gradually.

A key difference between Fantail and another popular preference learning package WEKA-LR is: Fantail uses the rank vector format (similar to the multi-target regression setting) rather than the order/explicit preference vector format. So in Fantail, label ranking is treated as a special case of the multi-target regression problem. The advantage of the Fantail approach is that both multi-target and label ranking algorithms can be used and tested under a unified framework.

Usage

See https://github.com/quansun/fantail-ml/blob/master/java/fantail-1-1-3/src/fantail/examples/LabelRankingSingleAlgoExample01.java for an example

Benchmark Datasets

/datasets/iris_.arff (an example dataset showing the data format used by Fantail)

A collection of 26 label ranking datasets can be downloaded from /datasets

Algorithms (8)

• AverageRanking (a baseline ranker)
• RankingWithkNN (based on a nearest neighbour algorithm)
• RankingWithBinaryPCT (based on predictive clustering tree for ranking)
• RankingByPairwiseComparison
• BinaryART (approximate ranking tree)
• ARTForests (approximate ranking tree forests)
• Label Ranking Tree (WEKA-LR's LRT, note: this algorithm has been removed from version 1-1-3)
• RankingViaRegression (multiple single-target regression)

Evaluation Metrics

• Spearman's rank correlation coefficient
• Kendall's Tau
• MAE
• RMSE

TODOs (major features)

• Curds and Whey Multivariate Responses
• Constraint Classification
• Bagging
• Boosting
• MetaRule Generator
• NDCG@X
• GUI/Visualisation
• 2D/3D permutation polytopes for rank data
• Experimenter

Citing Fantail

If you want to refer to Fantail in a publication, please cite the following paper:

Quan Sun and Bernhard Pfahringer. Pairwise Meta-Rules for Better Meta-Learning-Based Algorithm Ranking. Machine Learning, 93(1):141-161, Springer US, 2013, DOI: 10.1007/s10994-013-5387-y

http://quansun.com/pubs/qs-mlj13.pdf

Many TODOs, so please send me an Email if you would like to contribute to Fantail!

Quan Sun [email protected]

10/2013