I know that the frozen graph was trained with v.0.1.1 and < tensorflow 1.4
If i use v.0.2.0 and tensorflow 1.6 and continue training from frozen graph, will this cause issues?
It should, except we have not yet a 0.2.0 model. So if you try to continue training from frozen graph on TensorFlow r1.6, I’m not sure it’s going to work. It might, though, could be interesting to get your feedback on this 
I have continued the 0.1.1 graph with 3 epoch with 1.6 and it seems over trained to the 1.6 data, i am not sure it that is what happened, but any files i throw at it, it tries to map to the spellings that were only in the new training set