run_bi-lstm-cnn-crf3.sh is missing

[MS-NLP]

Thanks for your contributing such a great work. It seems that run_bi-lstm-cnn-crf3.sh is different from the official run_bi-lstm-cnn-crf.sh, and the former is missing in the branch. Could you please also share it? Thanks.

[SteffenEger]

Thanks for your interest! It's a slight modification of the script you find in the official branch that basically allows for outputting the predictions, etc. I'll try to upload it in the coming days.

[SteffenEger]

Hi there, sorry for the delay, but now I uploaded some of the modifications we made to the original code, see acl2017.tgz. Hope it works, -Steffen

p.s. There's one flag that you can change if you want. It's after the line "# MODEL SELECTION ON DEV CRITERION". It states whether you want to use loss or F1 as a stopping criterion on the dev set.

[MS-NLP]

Many thanks for your hard work. I will investigate it.

[tuhinjubcse]

Hi Steffen

Thank you so much for your help . It seems like the run_bi-lstm-cnn-crf3.py has multiple indentation issues. Can you please please look into it . I am afraid if i try to fix on my own n misunderstand anything , it ll be very hard to debug considering the code is so big now

[SteffenEger]

Hi there,

If it has indentation issues, probably best is to use python2.7 - that's most likely also what we used at the time. That said, instead of using BILSTM-CNN-CRF from Ma & Hovy, you could also use the code from Lample et al. (Neural Networks for Named Entity Recognition), the code from Reimers & Gurevych (Reporting score distributions makes a difference), or others. The underlying model is the same.