How do we want to add DeepLearning capability?

[Koji Kawamura <ij...@apache.org>]

Hi all,

A PR is submitted to add to do classification or prediction using a
pre-built model with DeepLearning4J library (thanks @mans2singh!).
https://github.com/apache/nifi/pull/2686

I found following things can/should be improved so that people can use
it easier from a NiFi flow:
- To utilize evaluation results at the downstream flow. Example flow
template is provided, however it requires external database.
(RecordLookup, enrichment pattern would be more appropriate?)
- Vectorize input data. (Record reader/writer, conversion?)

To me, the key point is "How easy and naturally NiFi user can use
DeepLearning within their NiFi flow?" A sophisticated DeepLearning
support may not be able to accomplish with just a single Processor.

I will keep reviewing PR2686, but anyone who is interested in, or has
some knowledge in Machine Learning or Deep Learning stuffs, please
join!

Thanks,
Koji