Hi there! this is my mini-repo on hyperbolic manifolds. Manifolds realized here are:
- Poincare Ball
- ...
At the moment the following modules are realized aat the basis of torch:
- Linear
- Embedding
- Hyperplane (module that outputs unnormalized logits)
- Wrapper for custom callable function to map them onto the manifold (called
HyperbolicWrapper) - ...
import torch
from torch import nn
from poincare_ball import PoincareBall
from modules import HyperbolicLinear, Hyperplane, HyperbolicParameter
from rsgd import RSGD
data = torch.rand(16, 25)
manifold = PoincareBall()
model = nn.Sequential(
nn.Linear(25, 10),
HyperbolicLinear(10, 5, manifold),
Hyperplane(5, 1, manifold)
)
euclidean_optimizer = torch.optim.SGD(
[p for p in model.parameters() if not isinstance(p, HyperbolicParameter)], lr=0.001
)
hyperbolic_optimizer = RSGD(
[p for p in model.parameters() if isinstance(p, HyperbolicParameter)], manifold, lr=0.001
)
euclidean_optimizer.zero_grad()
hyperbolic_optimizer.zero_grad()
logits = model(data)
loss = logits.mean()
loss.backward()
euclidean_optimizer.step()
hyperbolic_optimizer.step()