qaq-v / hetgt Goto Github PK

您好，我想请问一下如何获取论文中提到的LDC2017T10和LDC2015E86数据集？似乎无法从开源渠道获取，贵组织是向LDC购买的嘛？

I test the en2cs dataset on your codes, and I met two issues.
First, when I use nmt_preprocess.py, the list index out of range error occurred. I checked out the data and found in some sentences, the number of dependency tags doesn't equal the number of tokens. It is indeed the error in the corpus. But did you meet this error when you did the experiments?
Second, when i use the preprocess.py, another error occurs, KeyError: 'f', when it runs to the code
dataset = inputters.Dataset( fields, readers=([src_reader, tgt_reader, grh_reader] if tgt_reader else [src_reader, grh_reader]), data=([("src", src_shard), ("tgt", tgt_shard), ('grh', grh_shard)] if tgt_reader else [("src", src_shard), ('grh', grh_shard)]), dirs=([opt.src_dir, None, None] # Cannot use _dir with TextDataReader if tgt_reader else [opt.src_dir, None]), sort_key=inputters.str2sortkey[opt.data_type], filter_pred=filter_pred ).
this is my first time using openNMT so it is a bit difficult to cope with the problem. For the sentences in the first issue, I deleted the questionable sentences, and keep the number of the sentences in EN equals that in CS. Could you help me cope with this problem?

For the fully-connected attention structure, does it mean fully connected between all tokens (both nodes and edge tokens) or just the node-edge tokens as shown in Figure 1?

Thanks for the artefact. I am interested to apply this library for a machine translation task. However, in my case, I have to feed a custom dataset.

Do you think is it possible with HetGT? Also interested to know have you attempted to apply HetGT on a machine translation task?