tticoin / LSTM-ER

Implementation of End-to-End Relation Extraction using LSTMs on Sequences and Tree Structures in ACL2016.
Apache License 2.0
224 stars 73 forks source link

Test loss is too huge #9

Open WakedupChan opened 7 years ago

WakedupChan commented 7 years ago

I trained the model on the sem-eval 2010 dataset without changing source code. however, while training, the test loss grows and even reaches to 41605.6, and the rel micro and macro PRF are 0. Any one can tell me why and how to fix it? Thanks very much!