ConMask
ConMask copied to clipboard
bxshi
ConMask model described in paper Open-world Knowledge Graph Completion.
Hi! Thanks for your code! I am trying to run the Closed and Open world evaluation on the Dbpedia50 dataset, following the instructions in README. However, I run into an...
If I want to create a new dataset. How can I get those files, such as vocab.txt, train.tails.values and eval.tails.values.closed?
Hi bxshi, when I run the dkrl.py, always raise ValueError: Cannot feed value of shape (2, 14904) for Tensor 'static/content_matrix/initial_value:0', which has shape '(14904,)' [https://github.com/bxshi/ConMask/blob/master/ndkgc/models/dkrl.py#L202](url) I can't find a correct...
Thanks for uploading code. I'm really interested in your work. My question is how to preprocess dbpedia datasets, because `descriptions.txt` contains, something like `Paris,_Texas 2056 paris texas city county seat...
Can I just use 2 GPU?OR reduce the batch size?
DATA_DIR = './data/fb15k/' model = DKRL( entity_file=os.path.join(DATA_DIR, 'entities.txt'), relation_file=os.path.join(DATA_DIR, 'relations.txt'), vocab_file=os.path.join(DATA_DIR, 'vocab.txt'), pretrain_vocab_file=os.path.join(DATA_DIR, 'glove.6B.100d.txt'), content_file=os.path.join(DATA_DIR, 'descriptions.txt'), train_file=os.path.join(DATA_DIR, 'train.txt'), valid_file=os.path.join(DATA_DIR, 'valid.txt'), test_file=os.path.join(DATA_DIR, 'test.txt'), all_triples_file=os.path.join(DATA_DIR, 'all_triples.txt') ) Can the above data set...
hello , bxshi: could you share the running time about the model of dkrl in your code, and i notice that in your paper, the statistic of dkrl is different...
In the main func, the arguments to tail_target_checker() contains **dic**. Inside this method, a new list of target entities are defined and top-10 ids are computed with respect to new...
