Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

For OCGNN models, update hypersphere radius, and filter training data with active_mask #95

Merged
merged 10 commits into from
Jan 3, 2024
Merged

For OCGNN models, update hypersphere radius, and filter training data with active_mask #95

Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
ce15c10
Update hypersphere radius, and filter training data with active_mask
ahmed3amerai Dec 15, 2023
77d754c
Update hypersphere radius, and filter training data with active_mask
ahmed3amerai Dec 15, 2023
2695273
Set default warmup as math.inf
ahmed3amerai Dec 15, 2023
4a36ac8
Updating other models
ahmed3amerai Dec 19, 2023
d061151
Merge branch 'main' into main
ahmed3amerai Dec 21, 2023
ecc2caa
Keep accepted changes and Revert rest
ahmed3amerai Jan 3, 2024
a134943
Keep accepted changes and Revert rest
ahmed3amerai Jan 3, 2024
b36165d
Keep accepted changes and Revert rest
ahmed3amerai Jan 3, 2024
76f3fb9
Keep accepted changes and Revert rest
ahmed3amerai Jan 3, 2024
b1d1d05
Cleaning the pull request
ahmed3amerai Jan 3, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 6 additions & 4 deletions pygod/nn/ocgnn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,17 +108,19 @@ def loss_func(self, emb):
score : torch.Tensor
Outlier scores of shape :math:`N` with gradients.
"""
if self.warmup > 0:
with torch.no_grad():
self.warmup -= 1
self.c = torch.mean(emb, 0)
self.c[(abs(self.c) < self.eps) & (self.c < 0)] = -self.eps
self.c[(abs(self.c) < self.eps) & (self.c > 0)] = self.eps

dist = torch.sum(torch.pow(emb - self.c, 2), 1)
score = dist - self.r ** 2
loss = self.r ** 2 + 1 / self.beta * torch.mean(torch.relu(score))

if self.warmup > 0:
with torch.no_grad():
self.warmup -= 1
self.r = torch.quantile(torch.sqrt(dist), 1 - self.beta)
self.c = torch.mean(emb, 0)
self.c[(abs(self.c) < self.eps) & (self.c < 0)] = -self.eps
self.c[(abs(self.c) < self.eps) & (self.c > 0)] = self.eps

return loss, score