wikidata

[huiwudiyi]

when I run runserver.py .I found a error in "questionanswering/grounding/graph_queries.py"

from wikidata import scheme, endpoint_access, queries
ImportError: cannot import name 'scheme'

I have already installed wikidata.

[daniilsorokin]

Hi,

some of our internal packages are still missing from the code. It was initially published to give more insights into the paper and the experiments.

We are planing with the missing parts to update it in October to support the training and running pre-trained models. Sorry for the inconvenience!

Best,
Daniil

[huiwudiyi]

ok ,look forward to your update!

[daniilsorokin]

Hi, I have finally added a lot of additional info on required internal packages and Wikidata endpoint. Will expand on it even further, but let me know if need anything specific.

[BettyHcZhang]

Hi，Thank you for your share. I have followed your instructions above. I download the zip of wikidata ,Unfortunately I get the error as below:
File "/home/coling2018-graph-neural-networks-question-answering/questionanswering/grounding/graph_queries.py", line 6, in
from wikidata import scheme, endpoint_access, queries
File "/home/coling2018-graph-neural-networks-question-answering/wikidata/endpoint_access.py", line 4, in
from SPARQLWrapper import SPARQLWrapper, JSON
ModuleNotFoundError: No module named 'SPARQLWrapper'

Actually,I just want to run through the model to see the entire workflow. Just for quick understand the paper better. In fact, I do not need the final model.

[daniilsorokin]

Hi,

try installing SPARQLWrapper with:

pip install SPARQLWrapper

Best,
Daniil

[BettyHcZhang]

Thank you very much!

[daniilsorokin]

Sure, get it here: https://public.ukp.informatik.tu-darmstadt.de/coling2018-graph-neural-networks-question-answering/webqsp.examples.train.silvergraphs.02-12.el.unittest.json

and put into data/generated/ folder.

[BettyHcZhang]

Thank you for your help. I have another question about how the GNN used in semantic graph.
Suppose that:
we use simple GNN model: f=f(AHW). A is adjacent matrix of nodes.H is embedding of labels. W is wights matrix. Vg=ReLU(Wh + b). Loss as your paper.
For one question , we have N candidate graphs.Then we have N Vg output.So we trained the wights matrix(W) by SGD on each f(AHW). Then We have N trained graph model.
Then , I have a new question, it has M candidate graphs.How to use previous trained model to get the right answer?Which model should I use in those N graph model?
Looking forward to your reply!

[daniilsorokin]

To answer your question: you do not train N graph models, you train a single model that is applied across all graphs.

[blackkingback]

Hey gusy,what's up!I try to run runserver.py,but I can not find the qaserver.pyt under questionanswering folder.Did I miss something?

[purvanshi]

@daniilsorokin Is there any update on a newer version of the code with required dependencies?

[Dinxin]

@daniilsorokin I have got this error: ImportError: cannot import name 'queries' from 'wikidata' (/root/anaconda3/envs/nlp/lib/python3.8/site-packages/wikidata/init.py)

So, how to install the python module named "wikidata"?

The command I adopted is "pip install wikidata".