Skip to content

Conversation

sgreenbury
Copy link
Collaborator

@sgreenbury sgreenbury commented Aug 25, 2025

Closes #750.

This PR:

  • Add delta method for calculating transformed mean and variances only (ignores covariance)
  • Add tests
  • Update default output_from_samples=False
  • Update dim reduction notebook to use method
  • Adds notebook comparing delta method with sampling approach performance and speed
  • Enables fast computation when transforms affine
  • Adds Jacobian caching when affine

Next steps:

  • Test implementation further and compare with sampling approach
  • Consider optional support for higher order terms for variance. Update: I think this could be left for a separate issue and would be good to motivate by a requirement for additional precision in the variance as will make the calculation more expensive (like for second order mean)
  • Consider supporting covariance for inputs and impact on accuracy
  • Add support for calculating mean only with second order terms
  • Add property to TransformedEmulator to indicate if all y transforms are affine
  • Consider adding support for caching Jacobian for delta method if all y transforms are affine
  • Add tests back to transformed emulator to support full covariance and output from samples paths

- Add delta_method with new submodule
- Add tests
- Update default `output_from_samples=False`
- Update dim reduction notebook to use method
Copy link

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.


Powered by ReviewNB

@sgreenbury sgreenbury linked an issue Aug 25, 2025 that may be closed by this pull request
@sgreenbury sgreenbury marked this pull request as ready for review September 4, 2025 17:26
@sgreenbury sgreenbury requested a review from radka-j September 5, 2025 12:57
- Readd test when `full_covariance=True` and `output_from_samples=True`
- Remove obsolete TODOs
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Improve efficiency of predict with transformed emulator
1 participant