genggengcss/ontozsl

ontozsl's Issues

pretrain_struc can't be run

Hello,

I am trying to run your model, but it is not working. When I try to follow your instructions, I get the following error:

>>> python code/OntoEncoder/pretrain_struc.py --dataset NELL --hidden_dim 300
Traceback (most recent call last):
  File "/.../code/OntoEncoder/pretrain_struc.py", line 18, in <module>
    from code import KGEModel
ImportError: cannot import name 'KGEModel' from 'code' (/.../lib/python3.9/code.py)

Running it from any of the folders in the repo, does not work. How do I fix this? :)

Thank you!

Best,
Filip

SemSize args value

Hello,
I'm trying to setup and run the OntoZSL for the AwA2 dataset. While in the training process I would like to know the default args value for 'SemSize' argument. Since there is no default value mentioned in the args.py file the code throws an error in the training process while initializing the generator.

Thank you.

Load relation embeddings of multimodal embeddings

for i, rel in id2rel.items():
# load embeddings
if rel in entity_embeddings_img and rel in entity_embeddings_txt:
rel_embed_txt = entity_embeddings_txt[rel]
rel_embed_img = entity_embeddings_img[rel]****
here in entity_embeddings_img[rel] we didn't find any embedding because in entity_embeddings_img we have embeddings of entities only. How do you get entity_embeddings_img[rel] here.

Can you provide the entity_55000.npy and relation_55000.npy in your script progress_structure_embed.py

The images for the training and testing classes are not from the usual imagenet2012 dataset, can you provide them so we can actually run the test?

Or can you tell us which data package are these image classes coming from so we can download them altogether since It's not very reasonable to download them separately from the imagenet website.