Support processing datasets from S3 directly which have already been processed in the desired format

[ragavvenkatesan]

Hi,

We are trying to support SM algorithms, where the user has multiple channels of processed data in S3 that is ready. The plan is to create a AmazonS3BaseEstimator class that will inherit from AmazonBaseEstimator class and will overwrite the record_set method. In this method, we will return a list of RecordSet objects. The _TrainingJob class already works with a dictionary of multiple channels of data. Therefore, we will overwrite the fit method in our class with data element being a dictionary of multiple channels of record objects.

User will have to supply in their fit call, a list of S3 uri's.

Thanks,
R

[ragavvenkatesan]

CC: @mvsusp @owen-t @orchidmajumder @vrkhare

[owen-t]

Hi Ragav,

I'd be happy to help here - it's not clear what your question is.

Please arrange a call or meeting with me.

Thanks,

Owen.