Is it possible to extract embeddings?

[abtawfik]

I am playing around with this library but am curious if it is possible to extract node level or graph-level embeddings? I looked at the methods attached to the evaluator but the best I could find was evaluator.neuralogic_model.neural_model.getAllWeights().toString() which returned a string of the weights for each level.

The goal here is to perform some clustering on the examples. I very much appreciate any response and would like to expand using his library a lot more and potentially present it to my team once I can wrap my head around it.

[LukasZahradnik]

Hello. Yes, it should be possible to extract all learnable parameters using the state_dict method on your model or evaluator (evaluator.state_dict()).

I realized it might be better to give it a more apparent name, so in the new version (0.1.7) that I just released, there is an alias parameters for state_dict (evaluator.parameters()) as well as fixed retrieving parameter values of one specific type - might be good to update your version.

With the state_dict/parameters method, you will get a dictionary with keys "weights" (which contains actual parameters' values) and "weight_names".

[abtawfik]

Well that was simple!
Thank you and I'd be remiss if I didn't also say how much I enjoyed the elegance of the lifted relational neural network framework.

[LukasZahradnik]

Glad you enjoyed it!