[SPARK-7008] An implement of Factorization Machines based on Scala and Spark MLlib.

[zhengruifeng]

An implementation of Factorization Machines based on Scala and Spark MLlib.
Factorization Machine is a kind of state-of-the-art machine learning algorithm for multi-linear regression, and is widely used in recommendation systems.
Factorization Machines algorithm and its C++ implement LibFM works well in recent years' recommendation competitions.

A FMModel consist of three parts:
an Intercept (optional),
an one-way interactions weights like other linear models (optional),
a numFactors * numFeatures matrix representing the factors of each feature (mandatory).

I implement the training algorithm with SGD provided by MLlib's GradientDescent (to use MLlib's SGD, the model is encoded to a dense vector during training).

Ref:
http://libfm.org/
http://doi.acm.org/10.1145/2168752.2168771
http://www.inf.uni-konstanz.de/~rendle/pdf/Rendle2010FM.pdf

[witgo]

Here should be deleted.

[AmplabJenkins]

Can one of the admins verify this patch?

[mengxr]

@zhengruifeng Thanks for submitting your impl as a Spark package at http://spark-packages.org/package/zhengruifeng/spark-libFM! As we discussed on the JIRA, we need some time to evaluate FM and discuss a scalable solution. So do you mind closing this PR for now?

[zhengruifeng]

OK, please close it.

On May 1, 2015, at 5:35 AM, Xiangrui Meng [email protected] wrote:

@zhengruifeng https://github.com/zhengruifeng Thanks for submitting your impl as a Spark package at http://spark-packages.org/package/zhengruifeng/spark-libFM http://spark-packages.org/package/zhengruifeng/spark-libFM! As we discussed on the JIRA, we need some time to evaluate FM and discuss a scalable solution. So do you mind closing this PR for now?

—
Reply to this email directly or view it on GitHub #5591 (comment).

[withDataman]

hello, why don't load FM model?