Backport #1592 (#1594)

docs/release-latest.rst

@@ -65,3 +65,4 @@ Version 1.7
 - Fixed bug where calling `set_figure_params` could block if IPython was installed, but not used. :pr:`1547` :smaller:`I Virshup`
 - :func:`~scanpy.pl.violin` no longer fails if `.raw` not present :pr:`1548` :smaller:`I Virshup`
 - :func:`~scanpy.pl.spatial` refactoring and better handling of spatial data :pr:`1512` :smaller:`G Palla`
+- :func:`scanpy.pp.pca` works with `chunked=True` again :pr:`1592` :smaller:`I Virshup`

scanpy/preprocessing/_pca.py

@@ -113,7 +113,7 @@ def pca(
              Explained variance, equivalent to the eigenvalues of the
              covariance matrix.
     """
-    start = logg.info(f'computing PCA')
+    logg_start = logg.info(f'computing PCA')
 
     # chunked calculation is not randomized, anyways
     if svd_solver in {'auto', 'randomized'} and not chunked:
@@ -239,7 +239,7 @@ def pca(
             adata.varm['PCs'] = pca_.components_.T
         adata.uns['pca']['variance'] = pca_.explained_variance_
         adata.uns['pca']['variance_ratio'] = pca_.explained_variance_ratio_
-        logg.info('    finished', time=start)
+        logg.info('    finished', time=logg_start)
         logg.debug(
             'and added\n'
             '    \'X_pca\', the PCA coordinates (adata.obs)\n'
@@ -249,7 +249,7 @@ def pca(
         )
         return adata if copy else None
     else:
-        logg.info('    finished', time=start)
+        logg.info('    finished', time=logg_start)
         if return_info:
             return (
                 X_pca,

scanpy/tests/test_pca.py

@@ -114,3 +114,27 @@ def test_pca_reproducible(pbmc3k_normalized, array_type, float_dtype):
     assert_equal(a, b)
     # Test that changing random seed changes result
     assert not np.array_equal(a.obsm["X_pca"], c.obsm["X_pca"])
+
+
+def test_pca_chunked(pbmc3k_normalized):
+    # https://github.com/theislab/scanpy/issues/1590
+    # But also a more general test
+
+    # Subsetting for speed of test
+    pbmc = pbmc3k_normalized[::6].copy()
+    pbmc.X = pbmc.X.astype(np.float64)
+    chunked = sc.pp.pca(pbmc3k_normalized, chunked=True, copy=True)
+    default = sc.pp.pca(pbmc3k_normalized, copy=True)
+
+    # Taking absolute value since sometimes dimensions are flipped
+    np.testing.assert_allclose(
+        np.abs(chunked.obsm["X_pca"]), np.abs(default.obsm["X_pca"])
+    )
+    np.testing.assert_allclose(np.abs(chunked.varm["PCs"]), np.abs(default.varm["PCs"]))
+    np.testing.assert_allclose(
+        np.abs(chunked.uns["pca"]["variance"]), np.abs(default.uns["pca"]["variance"])
+    )
+    np.testing.assert_allclose(
+        np.abs(chunked.uns["pca"]["variance_ratio"]),
+        np.abs(default.uns["pca"]["variance_ratio"]),
+    )