reproducible experiments

[lzw950905]

Describe the bug
I want to use the experiment setting published in ''Bringing Light Into the Dark: A Large-scale Evaluation of Knowledge Graph Embedding Models Under a Unified Framework.''. One setting is "training_loop": "OWA". But when I run, it shows that ''ValueError: Invalid TrainingLoop name: OWA. Valid choices are: ['lcwa', 'slcwa']''.

To Reproduce
Steps to reproduce the behavior:

1. Use the pipeline and correponding para setting.
2. Run the code

Expected behavior
Run normally and could get the performance like you did after experiments.

[mali-git]

Hi @lzw950905,
We have updated the nomenclature in the subsequent versions. owa has been renamed to slcwa (stochastic local closed world assumption).

[cthoyt]

@mali-git shouldn’t we update the config files to reflect the new name?

[mali-git]

@mali-git shouldn’t we update the config files to reflect the new name?

Maybe we can add a conversion script (in the benchmarking repository)? This would ensure that the configurations can be executed with the version used for the experiments and ensures that the configurations can be used with the current version.