Adding a conda environment and migrating to tf 1.x

[lhk]

It's not easy to execute the code locally. There have been some breaking API changes and the code only works with specific versions of the imported packages.

I've migrated the code to tensorflow 1.0 using the official google migration script.
And I've created a conda environment that makes it possible to execute this code.

I'm not sure whether this introduces subtle bugs somewhere, but on my machine the training finishes and reports:

Epoch 39	Iter 27000	Loss 0.09313946131947733
train: Accuracy(>0.5): 0.9998210450966356, f1: 0.999910514541387, auc: N/A
dev: Accuracy(>0.5): 0.798, f1: 0.7589498806682577, auc: 0.9113439999999999