Validation split option

[johnml1135]

Machine.py should allow a validation split to be made. This should be passed in as a parameter but should default to off. We may want a Bleu score out of this - or an expectation as to how many steps we should use.

[ddaspit]

We don't have a specific requirement for this, so I am moving it to the backlog.

[johnml1135]

While the filtering may change, I think we can make an initial stab at this. Specifically, I am assuming that:

• We will have the same type of filter for validation as we have for training in general
• We will be able to select a percentage or a number, but default to something (500 segments?)
• We will fail if the number of segments for training is less than the number used for validation
• We will do the validation split in Machine (not in python)
• Machine.py will need to be updated to handle the validation split
• We will need to have a data format to get the data back on the S3 bucket to machine