Closed shaoyx closed 6 years ago
Hi, @mana-ysh I doubt there are some mistakes in the implementation of TransE when computing the gradient. It seems that the gradients in a mini-batch are not averaged by the batch_size.
https://github.com/mana-ysh/knowledge-graph-embeddings/blob/master/src/models/transe.py
Can you give some reason about such implementation?
Thanks!
Hi, @mana-ysh I doubt there are some mistakes in the implementation of TransE when computing the gradient. It seems that the gradients in a mini-batch are not averaged by the batch_size.
https://github.com/mana-ysh/knowledge-graph-embeddings/blob/master/src/models/transe.py
Can you give some reason about such implementation?
Thanks!