When I create a ConvLearner and do learn.predict(), I get an array of log-probabilities for the different classes. This means that if I calculate
I get an array full of ones. However, if I calculate the same expression for an RNN_Learner, I get an array of values different from 1. So how should I understand the outputs of learn.predict() for the RNN_learner? Maybe relatedly, if I calculate accuracy like
accuracy_np(learn.predict(), val_y), I get the same value as printed out in the final epoch of training for the ConvLearner, but not for the RNN_Learner.