search

ttrouill / complex

Source code for experiments in the papers "Complex Embeddings for Simple Link Prediction" (ICML 2016) and "Knowledge Graph Completion via Complex Tensor Factorization" (JMLR 2017).
Other
319 stars 83 forks source link

about the appendix A and B of the essay #4

Closed Misoknisky closed 6 years ago

Misoknisky commented 6 years ago

There is no trace about appendix A and appendix B of the Complex Embeddings for Simple Link Prediction

ttrouill commented 6 years ago

Hello, Indeed Appendix A describes the full SGD algorithm with gradients and all, this implementation is theano-based and hence uses auto-differentiation, that's why you'll see no trace of gradients in this code: it only needs to know the loss. Appendix A is here to help people implement the model by themselves, it is not a description of this implementation. PCA plots of Appendix B are not provided here, if you want to reproduce them you can easily access the real (r1) and imaginary (r2) parts of the relation embeddings once the model is trained in wn18_run.py via:

word_embs_exp.models["Complex_Logistic_Model"][0].r1.get_value(borrow=True)
word_embs_exp.models["Complex_Logistic_Model"][0].r2.get_value(borrow=True)

and then use your favorite PCA tool to visualize them.

Misoknisky commented 6 years ago

Thanks for your patience!!!!