nusnlp / ptrnetdecoding4jere Goto Github PK

I wan to try with 100d word2vectors, but dont have NYT29 corpus. I tried other 100d embeddings, the performance was bad. Could you please support 100d NYT29 pre-trained word2vectors?

Hi,
I use the NYT24 dataset and code you published, and the best result obtained is 0.693, which is very different from the result of 0.817 in your paper.
I want to know how to get the performance in your paper. Is the hyperparameter inappropriate? Or other reasons?

thank you.

Hello,
The number of "Entity overlap type" I tested on NYT24 is inconsistent with the data you gave in "Table 2: Statistics of train/test split of the two datasets.". Could you please provide us with your test script? .

The performance of HRL can't achieve so high, 66.4 reported in original paper.
Is the evaluation method not different from that paper?

Hello, thank you for your beautiful article. I run the following error code when running. What is the cause and how can I fix it?
line 948, in
os.environ['CUDA_VISIBLE_DEVICES'] = sys.argv[1]

IndexError: list index out of range

Hi, how to get adj_mat?

Hi!
I noticed that your train data is exactly the same as the one provided by Takanobu et al. (2019), but it seems you did not split the training data to generate dev data. So can you tell me where the dev data comes from? Thanks!

I found that you calculated the performance of relation and entity generation in your paper. But the code you released seems doesn't contain this part. I think the methods you use are different from Zeng's(released), I want to know how to calculate them. Otherwise, will you release more detailed evaluation codes?

at Line 887 of word_decoder.py ,the outputs' dimention is [batch_size*sequence_length , vocab_size], but the target's dimention is [batch_size, sequence_length]? How to calculate the loss?

Hi! I found you use different evaluation metric on dev data and test data.
Won't this cause the problem of generalization?
Why not use cal_f1() method of test data on dev data?

Thanks for your excellent work!
When I run the command with "python word_decoder.py 0 1023 NYT29/ NYT29/word_decode_model train", it has an error about recordclass module. Did you have this trouble before? The error output looks like this:

Traceback (most recent call last):
File "word_decoder.py", line 10, in
from recordclass import recordclass
File "/home/fintech/anaconda3/envs/py35/lib/python3.5/site-packages/recordclass/init.py", line 24, in
from .datatype import datatype, clsconfig
File "/home/fintech/anaconda3/envs/py35/lib/python3.5/site-packages/recordclass/datatype.py", line 208
return f'{typename}({args}, **{kw})'
^
SyntaxError: invalid syntax

Nice work. I have two questions, first "What's the performance of GCN encoder ?"; second "In dataset folder, it exist *.dep files with adj_mat information, Where are they from?"