not sure what data should be placed on ../data #3

Activity

1. pasky commented on Feb 17, 2016

   

   pasky

   on Feb 17, 2016 via email

   Member

   More actions

   I'll be happy to help, but can you please elaborate? What are you trying to do? Do you want to retrain some YodaQA classifiers according to the instructions in YodaQA, or are you interested in just trying out a sentence similarity model? (If the latter, please take a look at https://github.com/brmson/dataset-sts - this is a legacy repository in that regard.)
2. hjk41 commented on Mar 24, 2016

   

   hjk41

   on Mar 24, 2016

   More actions

   I am trying to reproduce the results in Yu's paper. What data should I put into data? I have downloaded the jacana files but it seems the data files are formatted as xml, not txt.
   Also, what is the list file for? I mean trainlist.p. Where can I get the file?
3. pasky commented on Mar 24, 2016

   

   pasky

   on Mar 24, 2016

   Member

   More actions

   Hi. I'd recommend using the dataset-sts framework instead for that, even though it doesn't have the word overlap logistic regression layer yet. (We're working on adding that, contributions welcome.) The advantage is that you can also easily try more complex neural models. It also contains the Jacana-originated dataset nicely preprocessed.

   @Silvicek, how did you convert the pseudo-xml files to the txt files in this repo?