Possible mistake in local aggregation

[fratajcz]

Hi!

First of all I wanted to thank you for you code, I know it is a bit older but it is still the most legible resource on hyperbolic GNNs that I could find. I am currently dissecting the mechanisms and re-implementing them for pytorch geometric so I would be very happy if you could take a minute of your time.

I encountered what I think is a mistake in the exponential mapping of the locally aggregated feature matrix (hyp_layers.py, line 145):

output = self.manifold.proj(self.manifold.expmap(x, support_t, c=self.c), c=self.c)

where x is the intransformed input features in hyperbolic space and support_t is the result of the neighborhood aggregation in euclidean space.

according to you implementation of expmap(), the first argument is the tensor that should be mapped and the second is the reference point. The way line 145 is written, the input features which are already in hyperbolic space are mapped into hyperbolic space again w.r.t the euclidean aggregated neighborhoods. Did you mean this the other way round? Then it would also line up with equation 9 in the paper.

A short confirmation would be very nice, Thanks in advance!

[Tai-Xinwei]

Hi! I am also confused about it? Have you solved it yet？ Thank you！

[a912289748]

wo yun,it is a same mistake for me,can you help the people and me ,