Cannot replicate the results mentioned in the paper
utkipatel opened this issue · comments
Based on the code, we tried to generate the data files and used that for training the given model.
But we were not able to replicate the results. Infact, the results for fill in the blank queries are coming around 25% only for us, which I guess is random(as there are 4 options for each blank). Can you say when are you planning to upload the documented code along with data-files format?
@utkipatel How can I download files:
category_summarize_1000.json
cid2rcid_1000.json
?
@utkipatel How can I download files:
category_summarize_1000.json
cid2rcid_1000.json
?
@donghaoye Even I could not find the files anywhere, so I studied the code and the dataset(polyvore), and came up with my own files which are similar to them...
As far as I understood,
--> cid2rcid_1000.json : has mapping from categoryid to the corresponding node index in the graph
--> category_summarize_1000.json: for each category, it has frequency of that category, and list of items that belongs to that category....
sorry for my late reply, we have upload all our data and you can re-implemented easily it now , i am not sure why your implementation does not get a good result. if you are in urgent, you can sent e-mail to me, may be i could give some suggestion about your re-implementation.
