Is it possible to do a transfer learning on a new similar QA dataset by using the checkpoint from [Link](https://github.com/unilight/R-NET-in-Tensorflow#trained-model-weights) as an initialization point?