About the answer prediction

Hi Christopher,

I still have a small question:

In Class BoundaryPrediction at docqa/nn/span_prediction.py, in function get_best_span(), I find you convert start_logits and end_logits to function best_span_from_bounds().

As far as I am concerned, we should convert tf.nn.log_softmax(start_logits) && tf.nn.log_softmax(end_logits) rather than start_logits&end_logits ?

Thanks a lot for your time!
YeDeming