Trying CompGCN on pharmkg dataset about neuralkg HOT 3 CLOSED

Thanks for your interest in NeuralKG. CompGCN is a graph convolutional framework that jointly embeds entities and relations in a knowledge graph. It uses the ConvE model as the decoder model; for that, we set the fixed convolutional layer and fully connected layer. Therefore, changing the embedding dimension of the compgcn convolutional layer will lead to the dimension mismatch between the encoder and decoder.
In the subsequent update, we will provide more adjustable hyper-parameters to make the decoder model match the compgcn convolutional layer with different dimensions.

from neuralkg.

Basically I don't understand why x has 200 columns and all_ent has 300 columns... Shouldn't they have the same number of columns? And why does it work if I leave the embedding dimension set to 100?

from neuralkg.

Many thanks.

from neuralkg.