search

UKPLab / elmo-bilstm-cnn-crf

BiLSTM-CNN-CRF architecture for sequence tagging using ELMo representations.
Apache License 2.0
389 stars 81 forks source link

Hyperparameter optimization #19

Open Mahmedturk opened 5 years ago

Mahmedturk commented 5 years ago

How are you optimizing hyperparameter space in this implementation? Grid search or random search or bayesian?

nreimers commented 5 years ago

I would recommend random search. See: http://www.jmlr.org/papers/volume13/bergstra12a/bergstra12a.pdf

Mahmedturk commented 5 years ago

Is it already implemented in this model? and what is the loss function of this model with CRF as a classifier?

nreimers commented 5 years ago

No, you would need to implement it by yourself.

For the loss function see line 51: https://github.com/UKPLab/elmo-bilstm-cnn-crf/blob/master/neuralnets/keraslayers/ChainCRF.py

Mahmedturk commented 5 years ago

okay.