Factorization Machine

[ChandraLingam]

Please fill out the form below.

System Information

• Framework (e.g. TensorFlow) / Algorithm (e.g. KMeans): SageMaker/Factorization Machine
• Framework Version:
• Python Version:
• CPU or GPU:
• Python SDK Version:
• Are you using a custom image:

Describe the problem

Is factorization machine designed to work only with binary features? Do we need to one hot encode all features? How are real-valued features handled?

Minimal repro / logs

General Query on algorithm supported

• Exact command to reproduce:

[yangaws]

Hi @ChandraLingam ,

Factorization Machines is not limited to using binary features. All features, binary or not, can be represented in sparse format as a list of vector_index:value pairs.

[ChandraLingam]

Hi @yangaws - Thank you for the clarification.

For Factorization Machines, SageMaker accepts only recordIO/protobuf formatted input.

What is the recommended process for converting a sparse matrix to recordio/protobuf format?

I used scikit OneHotEncoder to encode my data and it returns a CSR Sparse matrix.

How do I convert this CSR Sparse Matrix efficiently into recordIO file?

[winstonaws]

One way to do so is to use the data conversion functions provided in sagemaker.amazon.common in the sagemaker python SDK. I think this should work: https://github.com/aws/sagemaker-python-sdk/blob/master/src/sagemaker/amazon/common.py#L113

[ChandraLingam]

Yes, I found out about that method after searching for it and tested it with my dataset. Thank you!

[Sandy4321]

may you share example how to use write_spmatrix_to_sparse_tensor?

[yangaws]

Hi @Sandy4321 ,

We have unit tests for this function here:
https://github.com/aws/sagemaker-python-sdk/blob/master/tests/unit/test_common.py#L66

These tests will illustrate basic usage of this function. Feel free to reply if you have further questions.

[yangaws]

@Sandy4321

Let's use #707 to track this issue. If you have more questions, please reply #707.

Thanks