Switch to a more sensible normalization strategy for loss metrics#70

Open

allevitan wants to merge 9 commits intomasterfrom

Commits on Mar 19, 2026

Attempted to update the normalizations for the loss functions in a way which allows for a nice normalization of the Poisson NLL. The slow tests will now likely fail
allevitan
committed
Added some better documentation to the loss functions
allevitan
committed
Get all tests passing by updating the loss thresholds for the new normalizations, and add the new normalization to all models
allevitan
committed
Add intensity_mse loss option to all models

allevitan
and
claude
committed
Update tutorial to use loss instance attribute pattern

allevitan
and
claude
committed
Update poisson_nll test to match sum-based implementation

allevitan
and
claude
committed
Add loss function coverage to slow reconstruction tests

allevitan
and
claude
committed
Restructure loss coverage in reconstruction tests

allevitan
and
claude
committed
Update the example near field ptycho code to use Poisson NLL
allevitan
committed