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

Add ScaleTensor #96

Open
wants to merge 10 commits into
base: feature/improve_loss_functions
Choose a base branch
from
Open

Add ScaleTensor #96

wants to merge 10 commits into from
+519 −21

Conversation

HCookie
Copy link
Member

@HCookie HCookie commented Oct 18, 2024 •
edited
Loading

ScaleTensor

  • Allows dynamic setup of scalars
  • Each scalar can be assigned to operate on a dimension or set of dimensions
  • Scalar is then resolved at scale time, allowing relative indices
    Examples
    --------
    >>> tensor = torch.randn(3, 4, 5)
    >>> scalars = ScaleTensor((0, torch.randn(3)), (1, torch.randn(4)))
    >>> scaled_tensor = scalars.scale(tensor)
    >>> scalars.get_scalar(tensor.ndim).shape
    torch.Size([3, 4, 1])
    >>> scalars.add_scalar(-1, torch.randn(5))
    >>> scalars.get_scalar(tensor.ndim).shape
    torch.Size([3, 4, 5])

HCookie and others added 3 commits October 18, 2024 15:28
@HCookie
Add ScaleTensor
72c91f5 
- Allows dynamic setup of scalars
- Rework variable_scaler to use it
@HCookie
Update ScaleTensor
83f7f5c 
- Consolidate usage of name
- Add subset_by_dim
- Add resolve
- Resolve ruff issues
@HCookie
Updates
e48d2c7 
- Fix docs
- Provide add_scalar in BaseWeighedLoss
@HCookie HCookie marked this pull request as ready for review October 18, 2024 18:25
@HCookie
Fix tests
6c6a1ab 
- Add scalar tests
@HCookie
Update loss function initialisation function
5d293ee 
- Allow a user to set `add_scalar_{key}`
@HCookie
Fix function name in loss functions
edc5ad1 
- scale_by_variable_scaling -> scale
@HCookie
Add iterator to ScaleTensor
56ac71f
@HCookie HCookie mentioned this pull request Oct 21, 2024
Open
@HCookie HCookie self-assigned this Oct 21, 2024
floriankrb
floriankrb reviewed Oct 21, 2024
View reviewed changes
src/anemoi/training/losses/weightedloss.py
Comment on lines +93 to +100
if len(self.scalar) == 0:
return x

if feature_indices is None:
return x * self.variable_scaling
return x * self.variable_scaling[..., feature_indices]
scalar = self.scalar.get_scalar(x.ndim)

if feature_indices is None or "variable_scaling" not in self.scalar:
return x * scalar
return x * scalar[..., feature_indices]
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't the piece of code be in the ScaleTensor class ?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Wdym? The scalar indexing? I would say it depends on the feature_indices, so it belongs in the loss function.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@floriankrb floriankrb floriankrb left review comments

@JesperDramsch JesperDramsch Awaiting requested review from JesperDramsch

@gmertes gmertes Awaiting requested review from gmertes

@sahahner sahahner Awaiting requested review from sahahner

Assignees

@HCookie HCookie

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@HCookie @floriankrb