search

facebookresearch / poincare-embeddings

PyTorch implementation of the NIPS-17 paper "Poincaré Embeddings for Learning Hierarchical Representations"
Other
1.67k stars 232 forks source link

euclidean embedding #68

Open HHalva opened 4 years ago

HHalva commented 4 years ago

What is the correct way of creating euclidean embedding with the code? There have been a few questions on this but didnt find an answer, and I would like to make a comparison in my upcoming paper. For now, what I have done is to add Euclidean to the list of embeddings manually (it isnt initially there) -- for now I have not made any changes to the learning rate.

lematt1991 commented 4 years ago

For now, what I have done is to add Euclidean to the list of embeddings manually

Yep, this should be enough to train with the euclidean manifold.

for now I have not made any changes to the learning rate.

You'll probably want to tune the learning rate.

HHalva commented 4 years ago

Great thanks!

HHalva commented 4 years ago

Doing this, I ran into the issues identified here [https://github.com/facebookresearch/poincare-embeddings/issues/35], would be very helpful to know what's going on.

In particular, is it a problem with how the metrics are calculated in the current code? Or is the training for euclidean space somehow broken? Or are the results in the NIPS'17 paper incorrect/very misleading?