Having some problems to train cg dataset with original code

[gbhong]

Hi, I tried to run your original codes to reproduce the training process in the train branch with cg dataset you've used.
However, I ran into the same type of codes for both basic and debug sessions.

When I try to pretrain the layers for ev (sh run/train.sh experiments/cg/basic/configs/train-ev.yaml), the following error occurs:

Traceback (most recent call last):
File "train.py", line 215, in
main()
File "train.py", line 202, in main
optimizer=optimizer)
File "/DeepEventMine/model/training.py", line 126, in train
epoch=epoch
File "/DeepEventMine/eval/evaluate.py", line 273, in eval
result_dir=result_dir)
File "/DeepEventMine/eval/evalEV.py", line 430, in evaluate_ev
params=params)
File "/DeepEventMine/eval/evalEV.py", line 77, in generate_events
for level, ev_preds_ in enumerate(ev_preds_levels_):
TypeError: 'NoneType' object is not iterable

When I looked into the output of ev, unlike outputs of ner and rel, it was like {"output": None}.

It seems like I need to save models for ner, rel, ev before sliding into the training process of end-to-end model.
But as I cannot save ev model in this process, I can't move on to the training of e2e model.

Thanks.