feat: stalign api and landmark alignment by selmanozleyen · Pull Request #1195 · scverse/squidpy

selmanozleyen · 2026-06-01T08:21:50Z

Continuation of #1162

codecov · 2026-06-11T08:39:19Z

Codecov Report

❌ Patch coverage is 36.15935% with 625 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.93%. Comparing base (dc82c73) to head (6916fd5).

Files with missing lines	Patch %	Lines
...idpy/experimental/tl/_align/_stalign_impl/_core.py	0.00%	215 Missing ⚠️
src/squidpy/experimental/tl/_align/_stalign.py	31.47%	134 Missing and 1 partial ⚠️
src/squidpy/experimental/im/_rasterize_points.py	11.45%	84 Missing and 1 partial ⚠️
src/squidpy/experimental/tl/_align/_api.py	55.15%	61 Missing and 13 partials ⚠️
...y/experimental/tl/_align/_stalign_impl/_helpers.py	0.00%	69 Missing ⚠️
src/squidpy/experimental/tl/_align/_io.py	36.50%	37 Missing and 3 partials ⚠️
src/squidpy/experimental/tl/_align/_landmark.py	87.09%	2 Missing and 2 partials ⚠️
src/squidpy/_validators.py	70.00%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1195      +/-   ##
==========================================
- Coverage   78.95%   74.93%   -4.03%     
==========================================
  Files          63       70       +7     
  Lines        9439    10417     +978     
  Branches     1574     1703     +129     
==========================================
+ Hits         7453     7806     +353     
- Misses       1469     2073     +604     
- Partials      517      538      +21

Files with missing lines	Coverage Δ
src/squidpy/types.py	`100.00% <100.00%> (ø)`
src/squidpy/_validators.py	`87.23% <70.00%> (-2.18%)`	⬇️
src/squidpy/experimental/tl/_align/_landmark.py	`87.09% <87.09%> (ø)`
src/squidpy/experimental/tl/_align/_io.py	`36.50% <36.50%> (ø)`
...y/experimental/tl/_align/_stalign_impl/_helpers.py	`0.00% <0.00%> (ø)`
src/squidpy/experimental/tl/_align/_api.py	`55.15% <55.15%> (ø)`
src/squidpy/experimental/im/_rasterize_points.py	`11.45% <11.45%> (ø)`
src/squidpy/experimental/tl/_align/_stalign.py	`31.47% <31.47%> (ø)`
...idpy/experimental/tl/_align/_stalign_impl/_core.py	`0.00% <0.00%> (ø)`

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

timtreis · 2026-06-16T13:53:19Z

+    radius = int(np.ceil(float(np.max(blur_values)) * 4.0))
+    denom = 2.0 * (dx * blur_values * 2.0) ** 2
+
+    for x_i, y_i in zip(x, y, strict=False):


How many points do we have? Should probably rasterize

wdym? This is the function for rasterization?

…direction - landmarks from shapes elements are read in their coordinate system, the frame the fit is registered onto, not the element's intrinsic one - key_added with shapes-element landmarks is refused up front, and the refusal that suggested it now suggests table_key + key_added - the volume path's initial_affine maps the volume onto the section, the reverse of the 2D paths; the docstrings say so - transform refuses a coordinate_system the fit was not made in

spatialdata is a hard dependency, so the importorskips never skipped; jax stays optional, so its skip stays.

rasterize_points takes only sdata positionally; the alignment functions keep their reference and query positional. The jitted solver loop takes only its first argument positionally.

…e-insensitive method and direction

… share the raster fit path, trim repeated docstrings

…wn frame or has none

…ntainers

Most of these had a second paragraph arguing for a design decision rather than telling a reader anything: why the axes are read off the transformation instead of taken as an argument, why JAX is imported lazily, why the frame check exists rather than being applied silently, why one flag beats two spellings. The fit subclasses each restated that they are a StalignFit and listed what they inherit, which the class line already says. What a caller cannot recover from the code stays: landmark units and row-order matching, the obsm frame rule, the bilinear-vs-nearest quantisation, the mass-conserving blur, the stop-gradient on the contrast M step, the upstream velocity-axis divergence, and why not jnp.atleast_3d. lddmm also repeated _lddmm_run's loop-fusion note and kept prose between Parameters and Returns, where numpydoc does not expect it. One sentence in _api.py's module docstring had lost its grammar.

for more information, see https://pre-commit.ci

The volume fit never passes landmarks, so the point term is zero and the value was dead; it existed only because `lddmm` required the argument. `sigmaP` is now optional there and checked only when points are given.

selmanozleyen mentioned this pull request Jun 1, 2026

Feat/gene imputation #1161

Draft

selmanozleyen requested a review from timtreis June 8, 2026 09:44