Hey there, thank you so much for open-sourcing this project.
From the current code, it looks like the hyperbolic embeddings are expected to be projected to/from a Poincaré disk with a curvature of 1.0. Trying out HoroPCA with embeddings projected to a different space curvature (e.g. 0.1) results in funky-looking plots. Would you have an idea on what types of change would be required to accommodate these extra curvature values ? From the code I see some comments on curvature parameters:
https://github.com/HazyResearch/HoroPCA/blob/c9a99967a62688711f6b63dab3baee53fe2d8430/geom/poincare.py#L15
Hey there, thank you so much for open-sourcing this project.
From the current code, it looks like the hyperbolic embeddings are expected to be projected to/from a Poincaré disk with a curvature of 1.0. Trying out HoroPCA with embeddings projected to a different space curvature (e.g. 0.1) results in funky-looking plots. Would you have an idea on what types of change would be required to accommodate these extra curvature values ? From the code I see some comments on curvature parameters: https://github.com/HazyResearch/HoroPCA/blob/c9a99967a62688711f6b63dab3baee53fe2d8430/geom/poincare.py#L15
Please let me know and thanks again! Darius