Does DSSTNE work on categorical string features/ binary features?

[priyankapatil811]

Hi,

I fed data with binary features and timestamp in a database timestamp format. The recs file was generated with no output, just 0s and some random numbers. Does DSSTNE accept only continuous numerical values?

Thanks!

[scottlegrand]

Could you provide as small sample of what the dataset looked like?

[slegrandA9]

Ping?

[scottlegrand]

Closing, reopen if there is still interest.

[beeva-enriqueotero]

I think it's just the opposite: amazon-dsstne doesn't accept numerical/analog values, only binary ("indicator" type). But maybe the problem is related to this hack that replaces values >10 with zeros

Related to https://github.com/amzn/amazon-dsstne/issues/66#issuecomment-260964095 and https://github.com/amzn/amazon-dsstne/issues/69

Regards

[scottlegrand]

Wow, nice catch! That needs to be fixed. That's completely unacceptable. I'm going to ask the origin of this hack to understand how to address it conclusively. Let me reiterate as the author the framework code that it does support both indicator and floating point inputs and outputs. What might not be here is the glue to transform the data properly, but I can supply code for doing so in that case.