Cannot overfit model with no CRF option

glample / tagger

Named Entity Recognition Tool

Apache License 2.0

1.16k stars 426 forks source link

Cannot overfit model with no CRF option #16

Closed dungtn closed 6 years ago

dungtn commented 8 years ago

Hi @glample,

I'm trying to overfit the LSTM without the CRF layer on a small dataset but the dev/test scores was about ~2%. I used the --crf 0 option, do I need to take any further steps?

chqiwang commented 8 years ago

What's your learning rate? Try use adadelta instead of sgd.

dungtn commented 7 years ago

It's because I didn't use the pre-trained skip100.