Skip to content

Mask past-only covariates during loss computation #379

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

Merged
merged 3 commits into from
Nov 17, 2025
Merged

Mask past-only covariates during loss computation #379

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9fd3278
Mask past-only covariates during loss computation
abdulfatir Nov 15, 2025
6fd9682
Clone task_future_target
abdulfatir Nov 17, 2025
b8032f1
Clone tensors when construcing slices
abdulfatir Nov 17, 2025
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
5 changes: 4 additions & 1 deletion src/chronos/chronos2/dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -477,6 +477,7 @@ def _construct_slice(self, task_idx: int) -> tuple[torch.Tensor, torch.Tensor |
task_n_covariates,
task_n_future_covariates,
) = self.tasks[task_idx]
task_past_tensor, task_future_tensor = task_past_tensor.clone(), task_future_tensor.clone()
task_n_past_only_covariates = task_n_covariates - task_n_future_covariates

full_length = task_past_tensor.shape[-1]
Expand All @@ -502,7 +503,9 @@ def _construct_slice(self, task_idx: int) -> tuple[torch.Tensor, torch.Tensor |
# the task_context_tensor by slicing the appropriate indices which we do below
if self.mode in [DatasetMode.TRAIN, DatasetMode.VALIDATION]:
# the first task_n_targets elements in task_context_tensor are the targets
task_future_target = task_past_tensor[:, slice_idx : slice_idx + self.prediction_length]
task_future_target = task_past_tensor[:, slice_idx : slice_idx + self.prediction_length].clone()
# mask out all rows corresponding to covariates
task_future_target[task_n_targets:] = torch.nan

if task_n_future_covariates > 0:
# the last task_n_future_covariates elements in task_context_tensor are the known covariates
Expand Down