Unable to reproduce the best result (F1: 0.58) on Weibo datasets?

LeeSureman / Batch_Parallel_LatticeLSTM

Chinese NER using Lattice LSTM. Reproduction for ACL 2018 paper.

129 stars 16 forks source link

Unable to reproduce the best result (F1: 0.58) on Weibo datasets? #24

Open iamlockelightning opened 3 years ago

iamlockelightning commented 3 years ago

Hi, thank you so much for you elegant reproduction of batched version of Lattice LSTM.

But I can't get the same results on Weibo Datasets, using the same emb\params\data. I just got F1 0.44 (normally 0.42-0.44), even when I tried different params.

Can you give some advice? & Does anyone can reproduce the best result?

iamlockelightning commented 3 years ago

@LeeSureman Pls. Look forward to your reply.