Spark-FFM

A Spark-based implementation of Field-Awared Factorization Machine. See http://www.csie.ntu.edu.tw/~cjlin/papers/ffm.pdf

The data should be formatted as

label field1:feat1:val1 field2:feat2:val2

to fit FFM, that is to extends LIBSVM data format by adding field information to each feature.

Currently, we support paralleledSGD and paralledAdagrad optimization methods, as they are more efficient in dealing with large dataset.

Besides, user can also choose to have FFMModel with/without global bias and one-way interactions.

Contact & Feedback

If you encounter bugs, feel free to submit an issue or pull request.