Probabilities of labels

[JustusJL]

Hello!

If I've understood LSTM and CRF correctly, then both of them are able to give an estimation of which label a word should get, for example: Word: hello No-Tag: 40% Name: 35% Location: 20% ID: 5%

I was wondering where in the code I can find this distribution over the labels where one label is chosen? I have the feeling that it's inside the prediction_step function in train.py, but as said, I'm not able to find it.

May someone please assist me with this?

Best regards, Justus