General DL

[kailuowang]

It seems that most of the existing utilities are all specialized at learning on image data. Is there any plan to support general deep learning against non-image data?
To be more specific, I would like to use a dnn to learn against a vast number of vectors (ideally retrieved directly from HIve). I also like to transfer learning, i.e. parameters between parts of different dnn architectures. Is supporting such tasks on the roadmap? If so I would love to contribute as well.

[allwefantasy]

1. Check this PR: NLP support #56 . It provides a TFTextTransformer which can convert text in dataframe to array-like: [num-words,num-features] and TFFileEstimator which provide ability to integrates tensorflow code
2. Transfer learning example is already provided in README.md.