fudannlp / neuralsentenceordering Goto Github PK

Can you provide the code for generating the pickle file from a dataset? Also the dataset seems to have been moved to a different location, please update the link.

Best regards.

Hi,

I've been trying to reproduce the results from your paper (https://arxiv.org/pdf/1607.06952.pdf) by running this code, but I can't get the same numbers at all.

I changed only this:

n_epochs = 100
sentence_modeling = 'LSTM'

The results that I'm getting are way off. After 1 epoch:

Test_score:
Top 1 beam
patial_correct_rate: 0.7842622396790232
total_correct_rate: 0.26833016906247176

After 50 epochs:

Test_score:
Top 1 beam
patial_correct_rate: 0.7364572575269629
total_correct_rate: 0.22077569839978303

Both metrics ("patial_correct_rate", which I understand is "Rouge-S" from the paper, and "total_correct_rate", which seems to be "P-all") are getting worse over time. Also, they are far from the numbers you reported: 0.82 for Rouge-S and 0.31-0.33 for P-all. How did you obtain those numbers? Is there any parameter that I should tweak?

Thanks for help,
Ștefan

Hi, i have already get the pickle file and the raw data, but can you provide the script to show how to generate the pickle? looking forward to your reply.
best regards.

How to get the dataset since the link given is not working.