feat(metrics): add smooth and baseline-corrected spectral weights by alxndrkalinin · Pull Request #34 · alxndrkalinin/cubic

alxndrkalinin · 2026-03-12T22:35:17Z

smooth_spectral_weights: SG-smoothed Wiener weights for spectral PCC
estimate_noise_baseline: frequency-dependent noise floor from running low-quantile of SG-smoothed log-power spectrum
baseline_snr2_weights: SNR²-based weights against freq-dependent baseline
spectral_pcc_baseline: new spectral PCC variant using baseline approach
spectral_pcc: add smooth/sg_window/sg_polyorder params for optional SG-smoothed weights

Summary by Sourcery

Introduce smoothed and baseline-corrected spectral weighting options for spectral PCC metrics.

New Features:

Add smooth_spectral_weights to compute Savitzky–Golay–smoothed Wiener-style spectral weights.
Add estimate_noise_baseline to derive a frequency-dependent noise floor from the smoothed power spectrum.
Add baseline_snr2_weights to compute SNR²-based spectral weights using the noise baseline.
Add spectral_pcc_baseline metric that applies baseline-corrected SNR² weights in Fourier-domain PCC.
Extend spectral_pcc to optionally use SG-smoothed spectral weights via new smoothing parameters.

Enhancements:

Improve flexibility of spectral PCC weighting by supporting optional smoothing parameters and alternative weight computation paths.

- smooth_spectral_weights: SG-smoothed Wiener weights for spectral PCC - estimate_noise_baseline: frequency-dependent noise floor from running low-quantile of SG-smoothed log-power spectrum - baseline_snr2_weights: SNR²-based weights against freq-dependent baseline - spectral_pcc_baseline: new spectral PCC variant using baseline approach - spectral_pcc: add smooth/sg_window/sg_polyorder params for optional SG-smoothed weights Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

sourcery-ai · 2026-03-12T22:35:32Z

Reviewer's Guide

Adds Savitzky–Golay-smoothed spectral weights and a new baseline-corrected spectral PCC metric, plus an option for spectral_pcc to use the smoothed weights instead of the existing ones.

File-Level Changes

Change	Details	Files
Introduce SG-smoothed Wiener-style spectral weights and optional use in spectral_pcc.	Add smooth_spectral_weights to compute Wiener weights from an SG-smoothed log-power spectrum with window-length clamping and optional high-frequency cutoff Extend spectral_pcc signature with smooth, sg_window, and sg_polyorder parameters and branch between existing spectral_weights and the new smooth_spectral_weights based on the smooth flag	`cubic/metrics/bandlimited.py`
Add frequency-dependent noise-baseline estimation and SNR²-based baseline weights, and use them to implement a new spectral_pcc_baseline variant.	Implement _running_quantile_1d to compute a sliding-window quantile used for baseline estimation Add estimate_noise_baseline to derive a smoothed power spectrum and monotone non-increasing noise baseline from the log-power spectrum using Savitzky–Golay smoothing and a running low quantile Add baseline_snr2_weights to compute SNR²-based radial weights with low-frequency exclusion, hard cutoff, and soft quantile-based capping Implement spectral_pcc_baseline, mirroring spectral_pcc’s pipeline but sourcing weights from estimate_noise_baseline and baseline_snr2_weights, supporting optional frozen radial weights, and reusing existing radial binning and apodization utilities	`cubic/metrics/bandlimited.py`

Tips and commands

Interacting with Sourcery

Trigger a new review: Comment @sourcery-ai review on the pull request.
Continue discussions: Reply directly to Sourcery's review comments.
Generate a GitHub issue from a review comment: Ask Sourcery to create an
issue from a review comment by replying to it. You can also reply to a
review comment with @sourcery-ai issue to create an issue from it.
Generate a pull request title: Write @sourcery-ai anywhere in the pull
request title to generate a title at any time. You can also comment
@sourcery-ai title on the pull request to (re-)generate the title at any time.
Generate a pull request summary: Write @sourcery-ai summary anywhere in
the pull request body to generate a PR summary at any time exactly where you
want it. You can also comment @sourcery-ai summary on the pull request to
(re-)generate the summary at any time.
Generate reviewer's guide: Comment @sourcery-ai guide on the pull
request to (re-)generate the reviewer's guide at any time.
Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
pull request to resolve all Sourcery comments. Useful if you've already
addressed all the comments and don't want to see them anymore.
Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
request to dismiss all existing Sourcery reviews. Especially useful if you
want to start fresh with a new review - don't forget to comment
@sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

Enable or disable review features such as the Sourcery-generated pull request
summary, the reviewer's guide, and others.
Change the review language.
Add, remove or edit custom review instructions.
Adjust other review settings.

Getting Help

Contact our support team for questions or feedback.
Visit our documentation for detailed guides and information.
Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

Implement random binomial splitting (Rieger et al., Optics Express 2024) as an alternative to checkerboard splitting for single-image FRC/FSC. Each pixel's photon count is split into n1 ~ Binomial(n, 0.5) and n2 = n - n1, producing two noise-independent images with preserved Poisson statistics and no spatial subsampling. New `split_type="binomial"` parameter threads through calculate_frc, frc_resolution, fsc_resolution, and calculate_sectioned_fsc. Supports counts mode (raw camera data with gain/offset/readout noise correction) and poisson_thinning mode (fallback for float/deconvolved images). Multi-repeat averaging via `n_repeats` produces correlation-std and resolution-std for uncertainty quantification. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add nbins_low parameter to zero out the lowest radial frequency bins before computing spectrally-weighted PCC, suppressing DC/illumination/ background contributions that can dominate the correlation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add frc_weights() to derive per-frequency reliability weights from single-image FRC, and spectral_pcc_frcw() which uses these weights for spectral PCC computation. Export both from cubic.metrics. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Copilot

Pull request overview

Adds new spectral weighting options and a binomial (photon-statistics-preserving) splitting path to improve robustness/interpretability of Fourier-domain correlation metrics (Spectral PCC, FRC/FSC), including uncertainty estimation via repeated splits.

Changes:

Introduces SG-smoothed and baseline-corrected spectral weighting utilities, plus new spectral PCC variants (baseline-based and FRC-weight-based).
Extends FRC/FSC preprocessing to support split_type="binomial" with optional read-noise correction and n_repeats aggregation (std estimates).
Adds binomial split implementation + unit tests, and updates spectral correlation data structures/docs accordingly.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
`cubic/metrics/bandlimited.py`	Adds smoothed/baseline/FRC-derived spectral weighting functions and extends `spectral_pcc` with optional smoothing + low-k exclusion.
`cubic/metrics/spectral/frc.py`	Adds binomial split support to FRC/FSC workflows, including repeat RNG handling and optional std outputs.
`cubic/image_utils.py`	Implements `binomial_split` (counts + poisson thinning modes) used by single-image FRC/FSC.
`cubic/metrics/spectral/analysis.py`	Extends result containers to include `correlation-std` and `resolution-std` fields.
`cubic/metrics/spectral/README.md`	Documents checkerboard vs binomial splitting and recommended usage patterns.
`cubic/metrics/__init__.py`	Exposes new bandlimited metrics helpers (`frc_weights`, `spectral_pcc_frcw`).
`tests/test_image_utils.py`	Adds unit tests covering statistical/behavioral properties of `binomial_split`.
`tests/metrics/spectral/test_frc.py`	Adds integration tests for binomial-split FRC/FSC behavior and repeat/stability expectations.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

- Use cubic.cuda abstraction (to_same_device/asnumpy) in binomial_split (C1) - Derive frequency scaling factor from image dimensions in frc_weights (C2) - Use nanmean/nanstd for multi-repeat FRC averaging (S3) - Add xy_std/z_std keys to single-pass binomial fsc_resolution (S4) - Make internal baseline functions private (_estimate_noise_baseline, etc) (S5) - Document resolution averaging asymmetry in FRC (S7) - Add ndim, gain, readout_noise_rms validation to binomial_split (N1) - Fix "1-bit" → "1/7" in frc_weights docstring (N2) - Replace assert with raise ValueError in frc_weights (N3) - Extract duplicate binomial log message to module constant (N4) - Fix checkerboard uncertainty in README table (N6) - Fix bin_delta type annotation to int in spectral_pcc_frcw (N7) - Add __all__ to bandlimited.py (N8) - Fix SG window fallback for tiny arrays in smooth_spectral_weights (Copilot) - Add frozen_weights length validation in spectral PCC functions (Copilot) - Fix gain convention docstring in binomial_split (Copilot) - Add nbins_low to spectral_pcc docstring (Copilot) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Replace xp/get_array_module pattern with to_same_device in all three spectral PCC functions (CLAUDE.md: prefer cubic.cuda abstractions) - Use cubic.scipy proxy for savgol_filter and median_filter instead of direct scipy imports (CLAUDE.md: device-agnostic wrappers) - Add xy_std/z_std keys to fsc_resolution mask backend return path to match the documented return contract - Warn when n_repeats>1 is silently ignored (checkerboard or two-image) - Clarify n_repeats docstring: requires single-image + binomial mode - Add tests for spectral_pcc(smooth=True) and nbins_low parameter Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add a procedural commit workflow checklist (list → group by concept → plan N commits → execute) to prevent grouping changes by workflow step. Simplify CLAUDE.md to just reference AGENTS.md. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add bandpass_spectral_pcc — a spectral PCC that operates within a frequency band defined by cumulative weight-mass percentiles. Uses w_i * n_i (weight × Fourier pixel count per ring) as the mass definition, with soft cosine tapers at band edges to avoid ringing. New public API: - percentile_band_taper: reusable band selection + taper helper that returns tapered weights and band diagnostics (k_low, k_high, k50, k90 in both normalized and physical units) - bandpass_spectral_pcc: full metric supporting three weight methods (simple, smooth_wiener, baseline_snr2) or pre-computed weights via frozen_weights. Returns float by default, (float, dict) with return_diagnostics=True. Also removes unused spacing parameter from spectral_pcc_frcw (FRC weights are inherently in index-frequency units). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

After mean subtraction, the DC bin power is ~0, mapping to log(1e-30) ≈ -69 — an extreme outlier that distorts the first few Savitzky-Golay fitted values. Replace DC with its neighbor before smoothing in both smooth_spectral_weights and _estimate_noise_baseline. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Extend spectral_pcc with a weighting parameter ("simple", "smooth_wiener", "snr2") to select the per-bin weight law, and taper_low for soft cosine low-frequency exclusion. The existing smooth=True is kept as deprecated sugar mapping to "smooth_wiener". Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

alxndrkalinin · 2026-03-18T17:08:22Z

Review of latest changes (`ec58ec1`, `12dc6df`)

Found 3 issues in the new `weighting`/`taper_low` additions to `spectral_pcc`:

~~`taper_low` ramp starts at exactly 0 — bin 0 gets hard-zeroed, no actual ramping~~

At index 0: `0.5 * (1 - cos(0)) = 0.0`, so `taper_low=1` gives a hard zero (no ramp). The `percentile_band_taper` function uses `arange(1, width+1)` to avoid this — `taper_low` should match that pattern for consistency.

Fixed in 7386979. Ramp now uses `arange(1, n+1)` matching `percentile_band_taper`. DC is already excluded by `bid=-1`.
~~`weighting` override logic cannot distinguish default from explicit `"simple"`~~

`weighting` defaults to `"simple"`, so `weighting != "simple"` is False even when the user didn't pass it. This means `smooth=True` always wins when `weighting` is left at default — but the docstring says `smooth` is "ignored when `weighting` is passed." Fix: use `weighting: str | None = None` as the default so explicit setting is detectable.

Fixed in 7386979. Default changed to `None` with proper sentinel logic.
~~No validation on `weighting` values — unknown strings silently fall through to `"simple"`~~

~~Unlike `bandpass_spectral_pcc` which validates against `_WEIGHT_METHODS`, passing `weighting="frc"` here silently uses the `else` branch (simple weights). Should raise `ValueError`.~~

Fixed in 7386979. Added `ValueError` for unknown values. Unified method names: `"baseline_snr2"` → `"snr2"` across both functions.

Minor: `weighting` should use `Literal["simple", "smooth_wiener", "snr2"]` type annotation for consistency with `bandpass_spectral_pcc`, and `smooth` deprecation should emit `warnings.warn(DeprecationWarning)`.

All items resolved in 7386979. `weighting` now uses `Literal[...] | None`.

Resolve conflicts: - image_utils.py: keep binomial_split + np.ndarray return type on label() - frc.py: keep feature branch refactored code, drop stale main-branch reverse-split block that was moved to _calculate_frc_single_pass Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ames - Fix taper_low ramp to use arange(1, n+1) matching percentile_band_taper (DC already excluded by bid=-1, bin 0 shouldn't be hard-zeroed) - Fix weighting default: None sentinel instead of "simple" so smooth=True backward compat works correctly - Add ValueError for unknown weighting values - Unify weight method names: "baseline_snr2" → "snr2" across both spectral_pcc and bandpass_spectral_pcc (one _WEIGHT_METHODS constant) - Add tests for weighting="snr2", unknown weighting, taper_low, and smooth=True backward compatibility - Fix bandpass_spectral_pcc test type narrowing for mypy Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Delete unused _spectral_pcc_baseline (~120 lines of dead code) - Change rng default from 42 to None in frc_weights and spectral_pcc_frcw (aligns with calculate_frc/frc_resolution) - Use Literal["checkerboard", "binomial"] for split_type parameter - Document n_repeats=3 default vs calculate_frc's n_repeats=1 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Replace assert with RuntimeError in _calculate_frc_single_pass for original_image1 None check (assert stripped by python -O) - Warn when n_repeats>1 is silently ignored on deprecated mask backend in fsc_resolution Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add 6 tests: - frc_weights: weight range [0,1], monotone envelope after nbins_low, square image requirement - spectral_pcc_frcw: identical images ~1.0, range [-1,1], shape mismatch All tests use explicit rng=42 for reproducibility. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Add smooth_spectral_weights to cubic/metrics/__init__.py (was in bandlimited.__all__ but missing from the package export) - Remove "(default)" from README splitting table headers — checkerboard is the default for calculate_frc/frc_resolution/fsc_resolution but frc_weights/spectral_pcc_frcw default to binomial Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add weighting="snr2" using global noise floor + SNR² formula (max(0, SNR-1)², matching the dynacell-validated Spectral_PCC_SNR2). Rename the existing frequency-dependent baseline variant to "snr2_baseline". Both available in spectral_pcc and bandpass_spectral_pcc. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

alxndrkalinin and others added 3 commits March 12, 2026 17:35

alxndrkalinin requested a review from Copilot March 13, 2026 00:48

Copilot started reviewing on behalf of alxndrkalinin March 13, 2026 00:49 View session

Copilot AI reviewed Mar 13, 2026

View reviewed changes

alxndrkalinin and others added 6 commits March 12, 2026 20:57

alxndrkalinin and others added 7 commits March 23, 2026 14:17

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(metrics): add smooth and baseline-corrected spectral weights#34

feat(metrics): add smooth and baseline-corrected spectral weights#34
alxndrkalinin wants to merge 17 commits into
mainfrom
feat/smooth-spectral-weights

alxndrkalinin commented Mar 12, 2026 •

edited by sourcery-ai Bot

Loading

Uh oh!

sourcery-ai Bot commented Mar 12, 2026 •

edited

Loading

Interacting with Sourcery

Customizing Your Experience

Getting Help

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

alxndrkalinin commented Mar 18, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

alxndrkalinin commented Mar 12, 2026 • edited by sourcery-ai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by Sourcery

Uh oh!

sourcery-ai Bot commented Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviewer's Guide

File-Level Changes

Interacting with Sourcery

Customizing Your Experience

Getting Help

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

alxndrkalinin commented Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review of latest changes (ec58ec1, 12dc6df)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

alxndrkalinin commented Mar 12, 2026 •

edited by sourcery-ai Bot

Loading

sourcery-ai Bot commented Mar 12, 2026 •

edited

Loading

alxndrkalinin commented Mar 18, 2026 •

edited

Loading

Review of latest changes (`ec58ec1`, `12dc6df`)