f score is -1 about ncrfpp HOT 3 CLOSED

So I have CoNLL-2003 dataset, which I have modified to be in the following format to be consistent with format you provided in the sample_dataset

EU I-ORG
rejects O
German I-MISC
call O
to O
boycott O
British I-MISC
lamb O
. O

Peter I-PER
Blackburn I-PER

(and more)

Now here is a screen shot of training after few iterations with the following configuration

batch_size = 10
embd : GloVe6B 100 dimensional

(other things same as your sample train config)

f1 score of dev and test set is consistently -1 throughout the training although the accuracy of train/dev/test keeps on improving. What exactly is the accuracy? how can increase in accuracy not help improve f1 score? Any clues?

from ncrfpp.

I changed the tagScheme so that it is in BIO mode, but the f-score is still -1 all throughout

from ncrfpp.

You use the wrong data format. Only BIO and BMES(BIOES) tag scheme are supported. You need to change your input data format rather than the tagScheme. The tagScheme is setted based on the input data format.

from ncrfpp.