Thank you very much for your work on the NERDA project, very much appreciated!
In the predict() function found in predictions.py you've added the functionality to return return_confidence which is super nice! However, when you extract the most likely tag using values, indices = outputs[i].max(dim=1) (from line 104) the prediction scores have not been put in a softmax function which would give confidence scores of each tag relative to the others.
It would be very nice to either
Being able to get all the tag probabilities for each prediction.
Hi @smaakage85,
Thank you very much for your work on the NERDA project, very much appreciated!
In the
predict()
function found in predictions.py you've added the functionality to returnreturn_confidence
which is super nice! However, when you extract the most likely tag usingvalues, indices = outputs[i].max(dim=1)
(from line 104) the prediction scores have not been put in a softmax function which would give confidence scores of each tag relative to the others.It would be very nice to either
predict()
or inforward()
(NERDANetwork) like they do in this basic pytorch tutorial.