herbertchen1

for example ，I got 88 types in Ontonote，not 89. {'/other/event': 0, '/other/food': 1, '/other/living_thing/animal': 2, '/other/internet': 3, '/organization/government': 4, '/location/structure/airport': 5, '/other/body_part': 6, '/location/transit/railway': 7, '/organization/transit': 8, '/other/legal': 9, '/organization/company':...

So we can not change the word embedding with the pretrained LM?

1

And training the LM is very hard...?

is there any reproducibility problem?

9

It seems many people think it's hard to got the result.....

with py2 torch0.3.1 break with an error

2

Traceback (most recent call last): File "/baseline/sentence-similarity/main.py", line 67, in runner.run(args.epochs, train_loader, dev_loader, test_loader, args.log_interval) File "/baseline/sentence-similarity/runners/__init__.py", line 76, in run trainer.run(train_loader, max_epochs=epochs) File "build/bdist.linux-x86_64/egg/ignite/engine/engine.py", line 223, in run File...

I run the shell script directly but cannot got the result on the paper

8

Only unseen=8.8357, seen=87.9500, h=16.0581, unseen class accuracy= 64.2205472255 May I ask which hyper parameter would influence the result so much?