limteng-rpi / fet Goto Github PK

why can't I find the relevant dataset

Thanks you for publish this work!

Can you upload datasets or the website of datasets?

Hi,

Following your guidance in README.md, I run this code on the OntoNotes datasets by:
python train.py --train /path/to/data/buffer/train --dev /path/to/data/buffer/dev
--test /path/to/data/buffer/test --ontology /path/to/data/type.txt --output /path/to/output
--elmo_option /path/to/data/eng.original.5.5b.json --elmo_weight /path/to/data/eng.original.5.5b.hdf5 --batch_size 200 --max_epoch 100 --eval_step 200 --gpu

But I can only obtain
best acc test: 54.25
best mac test: 72.94
best mic test: 67.68

Is there something ignored by me?
Looking forward to your reply.

Best.

Hi, I notice that your input data format is self-defined. Could you release your training data and the arguments you use for training?

I tried to use Bert embedding(dim: 768) in this task. And find out CUDA out of memory error in the first epoch even batch size was set to 3.
May I ask how much GPU memory need to be allocated when you use elmo embeddings and what is the common batch_size you usually use in this task?

Thanks a lot!

Hi

how did you convert the ontonote data to JSON format you mentioned in section 1, which script did this. same question for other dataset

I am new to this area, confused with data file in ontonote, thanks for your help!

best wish

Hello, I hope to compare with your method on FIGER(Gold), but I don't know how you divide the training set and validation set. Could you please share your processed data?