turboLJY/Few-Shot-KG2Text

How to obtain the two pretrained embedding files in preprocess.py

Opened this issue · 4 comments

LittleRedLynn commented

Sorry I'm a new guy in KG-to-Text task. One thing confused me is that I've already downloaded the pretrained embeddings provided in OpenKE website as you guided in ReadME, however, the files are all .bin and .txt format, seems they provide nothing related to .pkl or .npy files shown in your source code, could you show me some clues? Should I do some format conversion by myself?

turboLJY commented

You should run TransE in OpenKE based on our datasets to obtain the entity embeddings.

LittleRedLynn commented

You should run TransE in OpenKE based on our datasets to obtain the entity embeddings.

Could you please be a little bit more specific? There need 3 files to run TransE training process, your code in preprocess folder seems not be able to produce the 'train2id.txt' OpenKE needs, should I construct it by myself using the 'train_precessed.json'?
Another question is what the heck are the transe_dict.pkl and transe_embedding.npy? By running the TransE training code I can only obtain the weights checkpoint of TransE model, what kinds of operation should I do to get the two files?
Finally, can you just provide the embedding files you utilized in your experiment? Gratefully thanks.

DhavalTaunk08 commented

You should run TransE in OpenKE based on our datasets to obtain the entity embeddings.

Can you please tell how to add our dataset in TransE

spwidodo commented

Hi
Thanks for sharing your great job..

I have same problem with @LittleRedLynn.
By running Transe . I just get transe.ckpt file. There is no transe_dict.pkl and transe_embedding.npy.
Any solution for this?

Thanks a lot