Multi-spectrum bandpower window function - #53
Merged
Conversation
Lift the single-spectrum guard on Fisher.get_bandpower_window_function so TEB/TQU runs can convolve per-ℓ theory into expected bandpowers. Both the binned Fisher and the per-ℓ Fisher are block-ordered spectrum-major, so Q becomes block-diagonal (np.kron with the identity) and is bit-identical to the previous Q when nspectra == 1. The window is now (nspectra·n_bins, nspectra·n_ell). Spectra.convolve_theory_for_inference accepts either a flat spectrum-major per-ℓ array or a label-keyed dict; the spectrum-label ordering is factored into a shared _spectrum_labels() helper reused by _make_convolve_theory. Prerequisite for the bandpower-shape / effective-ell work (ADR-0019), whose effective multipoles differ per spectrum and so cannot be served for TEB without this. Closes #49
There was a problem hiding this comment.
Pull request overview
This PR extends CosmoForge’s QML inference utilities to support multi-spectrum (e.g., TEB/TQU) workflows when mapping per-ℓ theory to expected binned bandpowers. It removes the prior single-spectrum limitation by making the bandpower window function and the per-ℓ theory convolution operate on spectrum-major, block-ordered flattened vectors (or a label-keyed dict), which is a prerequisite for ADR-0019 follow-on work (effective multipoles per spectrum).
Changes:
- Generalizes
Fisher.get_bandpower_window_function()to return a(nspectra·n_bins, nspectra·n_ell)window fornspectra > 1using a block-diagonal sum-over-ℓ operator. - Extends
Spectra.convolve_theory_for_inference()to accept either a flat(nspectra·n_ell,)spectrum-major vector or a label-keyed dict of per-ℓ theory arrays. - Adds multi-spectrum tests validating window shape/linearity/normalization and dict-vs-flat convolution equivalence, and updates docs + changelog accordingly.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
src/cosmoforge.qube/qube/fisher.py |
Lifts the single-spectrum guard and builds a block-diagonal Q so the bandpower window supports nspectra > 1. |
src/cosmoforge.qube/qube/spectra.py |
Adds spectrum-label resolution helper and extends per-ℓ theory convolution to accept multi-spectrum flat vectors or label-keyed dicts. |
src/cosmoforge.qube/tests/test_binning.py |
Adds multi-spectrum unit tests for the bandpower window function and dict-input convolution path. |
docs/source/tutorials/basic_usage.rst |
Updates tutorial guidance to reflect multi-spectrum support in convolve_theory_for_inference(). |
docs/source/changelog.rst |
Adds an Unreleased changelog entry describing the new multi-spectrum window + theory-convolution inputs. |
Comment on lines
+1683
to
+1684
| cl_theory : np.ndarray or dict | ||
| Per-ℓ theory C_ℓ. Accepted formats: |
Owner
Author
There was a problem hiding this comment.
Fixed in 79580ce — parameter annotated np.ndarray | dict.
Comment on lines
+1753
to
+1758
| def _to_perell(arr: np.ndarray, what: str) -> np.ndarray: | ||
| """Trim one per-ℓ array to ℓ=lmin..lmax (accepts n_ell or lmax+1).""" | ||
| if arr.size == n_ell: | ||
| return arr | ||
| if arr.size >= lmax + 1: | ||
| return arr[lmin : lmax + 1] |
Owner
Author
There was a problem hiding this comment.
Fixed in 79580ce — added an ndim != 1 guard in _to_perell so a size-matching 2D entry (e.g. (n_ell, 1)) raises early; covered by test_convolve_theory_dict_rejects_2d_entry.
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #53 +/- ##
=========================================
Coverage ? 92.14%
=========================================
Files ? 5
Lines ? 1082
Branches ? 0
=========================================
Hits ? 997
Misses ? 85
Partials ? 0
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
Annotate the parameter as np.ndarray | dict to match the accepted forms, and reject non-1D per-ℓ arrays inside the dict path: a (n_ell, 1) entry whose size matches would previously slip past the size check and corrupt the spectrum-major concatenation. The ndim guard lives in _to_perell so every array routed through it is covered.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
Lifts the single-spectrum restriction on
Fisher.get_bandpower_window_functionso multi-spectrum (TEB/TQU) runs can convolve per-ℓ theory into expected bandpowers. Both the binned Fisher and the per-ℓ Fisher are block-ordered spectrum-major, so the sum-over-ℓ operatorQbecomes block-diagonal (np.kronwith the identity) — bit-identical to the previousQwhennspectra == 1. This is the prerequisite for the bandpower-shape / effective-ell work (ADR-0019), whose effective multipoles differ per spectrum and therefore cannot be served for TEB without it.Public surface
Fisher.get_bandpower_window_function()no longer raisesNotImplementedErrorfornspectra > 1; returns a window of shape(nspectra·n_bins, nspectra·n_ell)(was(n_bins, n_ell), still that for a single spectrum). Rows/cols are block-ordered spectrum-major, matchingget_bandpower_slices.Spectra.convolve_theory_for_inference()now accepts a flat(nspectra·n_ell,)per-ℓ array or a label-keyed dict{"TT": …, "EE": …}of per-ℓ theory. Single-spectrum output is numerically unchanged.Checklist
docs/source/changelog.rstupdated under Unreleased (Added).CONTEXT.mdupdated if this introduces or sharpens a domain term — no new term.basic_usage.rstno longer calls the method single-spectrum-only; sphinx build clean (no new warnings).uv run pytest src/cosmoforge.qube/tests/ -s→ 178 passed, 8 skipped).