uclanlp / synpg Goto Github PK

Hi, when I run train.syng.sh, I get an error:

Traceback (most recent call last):
File "train_synpg.py", line 285, in
train(epoch, model, train_data, valid_data, train_loader, valid_loader, optimizer, criterion, dictionary, bpe, args)
File "train_synpg.py", line 90, in train
sent_ = bpe.segment(sent_).split()
File "/home/dingp/synpg-master/subwordnmt/apply_bpe.py", line 59, in segment
new_word = [out for segment in self._isolate_glossaries(word)
File "/home/dingp/synpg-master/subwordnmt/apply_bpe.py", line 60, in
for out in encode(segment,
File "/home/dingp/synpg-master/subwordnmt/apply_bpe.py", line 144, in encode
word = tuple(orig[:-1]) + ( orig[-1] + '',)
TypeError: unsupported operand type(s) for +: 'int' and 'str'

It seems something wrong with the BPE? Could you help me ? thanks !

Hi,
We want to use a new dataset to test on this model. Could you open source your preprocessing code? More specifically, how to generate h5 files contaning sentence and syntax information from files processed by StanfordCoreNLP.
Big Thanks!

Hi,

Could you please provide the command you used to generate the parses to use as input, as well as any other preprocessing steps required? I have some data that I would like to test your system on, how should I generate an input file in the correct format?

Thanks!

Following the instructions for Training using: train_synpg.sh with the downloaded data

==== start training ====
Traceback (most recent call last):
  File "train_synpg.py", line 285, in <module>
    train(epoch, model, train_data, valid_data, train_loader, valid_loader, optimizer, criterion, dictionary, bpe, args)
  File "train_synpg.py", line 90, in train
    sent_ = bpe.segment(sent_).split()
  File "/home/wmk/synpg/subwordnmt/apply_bpe.py", line 59, in segment
    new_word = [out for segment in self._isolate_glossaries(word)
  File "/home/wmk/synpg/subwordnmt/apply_bpe.py", line 67, in <listcomp>
    self.glossaries)]
  File "/home/wmk/synpg/subwordnmt/apply_bpe.py", line 144, in encode
    word = tuple(orig[:-1]) + ( orig[-1] + '</w>',)
TypeError: unsupported operand type(s) for +: 'int' and 'str'

Following the instructions for Training using: train_parse_generator.sh

==== loading data ====
number of train examples: 45377426
number of valid examples: 12800
==== start training ====
Traceback (most recent call last):
  File "train_parse_generator.py", line 282, in <module>
    train(epoch, model, train_data, valid_data, train_loader, valid_loader, optimizer, criterion, dictionary, args)
  File "train_parse_generator.py", line 81, in train
    synt_ = ParentedTree.fromstring(synt_)
  File "/opt/conda/envs/synpg/lib/python3.7/site-packages/nltk/tree.py", line 669, in fromstring
    for match in token_re.finditer(s):
TypeError: cannot use a string pattern on a bytes-like object

The generation script gives 4 candidates outputs for each input - what is the "correct" way to select the preferred top-1 output? Is it simply the first generated output?

Hi, nice job! I find a bug in generate.py:

Line 90,

eos_pos = eos_pos[0]+1 if len(eos_pos) > 0 else len(idx)

the 'idx' is not defined.

Please can you add the license to the code. thanks!

Hello, I am in China. When I click the link "data" to download the data, I always get a download error when it reaches 500M. Could you please provide the original webpage for downloading data, and then I can download it from the corresponding webpage, or can you provide the link of Baidu net disk?