uma-pi1 / kge

LibKGE - A knowledge graph embedding library for reproducible research
MIT License
759 stars 124 forks source link

Number of model parameters. #142

Closed Demirrr closed 3 years ago

Demirrr commented 3 years ago

Hey,

thank you for great work! I was wondering whether the number of parameters of models --including RESCAL, ComplEx and ConvE have been reported in somewhere.

Cheers, Caglar Demir

rgemulla commented 3 years ago

It's printed and logged whenever you train a model. For the published models, run

kge dump checkpoint fb15k-237-rescal.pt --keys valid_trace | grep num_parameters | head -1

to extract it. Is this what you mean?

Demirrr commented 3 years ago

I am afraid something is incomplete.

  1. All steps in quick-start are completed, i.e., the script of toy example was successfully executed.

  2. Executing kge dump checkpoint fb15k-237-rescal.pt --keys valid_trace | grep num_parameters | head -1 throwed the following exception Exception: Could not find checkpoint in fb15k-237-rescal.pt

My main goal is to fully make use of the results reported in "YOU CAN TEACH AN OLD DOG NEW TRICKS! ON TRAINING KNOWLEDGE GRAPH EMBEDDINGS". More specifically, I am interesting in the number of trainable parameters for all models in the Table2. Apart from this, I was wondering whether you could name one or two frameworks that remove unseen entities from the test set.

Cheers

rgemulla commented 3 years ago

You need to download the respective checkpoint from the KGE homepage first. The checkpoint links are in the "pretrained model" column.

Demirrr commented 3 years ago

It worked like a charm. Thanks!