Update dev branch (scverse#2421)

* Fix anndata-dev install

* Fix test_ingest by setting X dtype explicitly

* Fix log1p test (implicit copy was no longer being made)

* Fix test that relied on implicit dtype conversion

* Fix dataset that now loaded in as float64 instead of float32

* Update visium default figure for matplotlib 3.7

* skip plotting test that changed on mpl 3.7 if mpl < 3.7 is installed

* Update release notes

.azure-pipelines.yml

@@ -50,7 +50,7 @@ jobs:
     displayName: 'Install dependencies'
 
   - script: |
-      'pip install -v "anndata[dev,test] @ git+https://github.com/scverse/anndata"'
+      pip install -v "anndata[dev,test] @ git+https://github.com/scverse/anndata"
     displayName: 'Install development anndata'
     condition: eq(variables['ANNDATA_DEV'], 'yes')
 

docs/release-notes/1.9.2.md

@@ -1,13 +1,9 @@
-### 1.9.2 {small}`the future`
-
-
-```{rubric} Documentation
-```
+### 1.9.2 {small}`2023-02-16`
 
 ```{rubric} Bug fixes
 ```
-- {func}`~scanpy.pp.highly_variable_genes` `layer` argument now works in tandem with `batches` {pr}`2302` {smaller}`D Schaumont`
-- {func}`~scanpy.pp.highly_variable_genes` with `flavor='cell_ranger'` now handles the case in {issue}`2230` where the number of calculated dispersions is less than `n_top_genes` {pr}`2231` {smaller}`L Zappia`
 
-```{rubric} Performance
-```
+* {func}`~scanpy.pp.highly_variable_genes` `layer` argument now works in tandem with `batches` {pr}`2302` {smaller}`D Schaumont`
+* {func}`~scanpy.pp.highly_variable_genes` with `flavor='cell_ranger'` now handles the case in {issue}`2230` where the number of calculated dispersions is less than `n_top_genes` {pr}`2231` {smaller}`L Zappia`
+* Fix compatibility with matplotlib 3.7 {pr}`2414` {smaller}`I Virshup` {smaller}`P Fisher`
+* Fix scrublet numpy matrix compatibility issue {pr}`2395` {smaller}`A Gayoso`

scanpy/datasets/_datasets.py

@@ -174,7 +174,8 @@ def paul15() -> ad.AnnData:
     backup_url = 'http://falexwolf.de/data/paul15.h5'
     _utils.check_presence_download(filename, backup_url)
     with h5py.File(filename, 'r') as f:
-        X = f['data.debatched'][()]
+        # Coercing to float32 for backwards compatibility
+        X = f['data.debatched'][()].astype(np.float32)
         gene_names = f['data.debatched_rownames'][()].astype(str)
         cell_names = f['data.debatched_colnames'][()].astype(str)
         clusters = f['cluster.id'][()].flatten().astype(int)
@@ -194,7 +195,7 @@ def paul15() -> ad.AnnData:
     # remove 10 corrupted gene names
     infogenes_names = np.intersect1d(infogenes_names, adata.var_names)
     # restrict data array to the 3461 informative genes
-    adata = adata[:, infogenes_names]
+    adata = adata[:, infogenes_names].copy()
     # usually we'd set the root cell to an arbitrary cell in the MEP cluster
     # adata.uns['iroot'] = np.flatnonzero(adata.obs['paul15_clusters'] == '7MEP')[0]
     # here, set the root cell as in Haghverdi et al. (2016)

scanpy/tests/test_embedding_plots.py

@@ -1,6 +1,7 @@
 from functools import partial
 from pathlib import Path
 
+import matplotlib as mpl
 import matplotlib.pyplot as plt
 from matplotlib.colors import Normalize
 from matplotlib.testing.compare import compare_images
@@ -304,6 +305,11 @@ def test_visium_circles(image_comparer):  # standard visium data
 
 
 def test_visium_default(image_comparer):  # default values
+    from packaging.version import parse as parse_version
+
+    if parse_version(mpl.__version__) < parse_version("3.7.0"):
+        pytest.xfail("Matplotlib 3.7.0+ required for this test")
+
     save_and_compare_images = image_comparer(ROOT, FIGS, tol=5)
     adata = sc.read_visium(HERE / '_data' / 'visium_data' / '1.0.0')
     adata.obs = adata.obs.astype({'array_row': 'str'})

scanpy/tests/test_ingest.py

@@ -17,10 +17,11 @@
         [7.0, 9.4, 6.8, 9.1, 8.0],
         [8.9, 8.6, 9.6, 1.0, 2.0],
         [6.5, 8.9, 2.2, 4.5, 8.9],
-    ]
+    ],
+    dtype=np.float32,
 )
 
-T = np.array([[2.0, 3.5, 4.0, 1.0, 4.7], [3.2, 2.0, 5.0, 5.0, 8.0]])
+T = np.array([[2.0, 3.5, 4.0, 1.0, 4.7], [3.2, 2.0, 5.0, 5.0, 8.0]], dtype=np.float32)
 
 
 @pytest.fixture

scanpy/tests/test_preprocessing.py

@@ -15,11 +15,11 @@
 
 
 def test_log1p(tmp_path):
-    A = np.random.rand(200, 10)
+    A = np.random.rand(200, 10).astype(np.float32)
     A_l = np.log1p(A)
-    ad = AnnData(A)
-    ad2 = AnnData(A)
-    ad3 = AnnData(A)
+    ad = AnnData(A.copy())
+    ad2 = AnnData(A.copy())
+    ad3 = AnnData(A.copy())
     ad3.filename = tmp_path / 'test.h5ad'
     sc.pp.log1p(ad)
     assert np.allclose(ad.X, A_l)
@@ -84,18 +84,19 @@ def test_mean_var_sparse():
 
 
 def test_normalize_per_cell():
-    adata = AnnData(np.array([[1, 0], [3, 0], [5, 6]]))
+    A = np.array([[1, 0], [3, 0], [5, 6]], dtype=np.float32)
+    adata = AnnData(A.copy())
     sc.pp.normalize_per_cell(adata, counts_per_cell_after=1, key_n_counts='n_counts2')
     assert adata.X.sum(axis=1).tolist() == [1.0, 1.0, 1.0]
     # now with copy option
-    adata = AnnData(np.array([[1, 0], [3, 0], [5, 6]]))
+    adata = AnnData(A.copy())
     # note that sc.pp.normalize_per_cell is also used in
     # pl.highest_expr_genes with parameter counts_per_cell_after=100
     adata_copy = sc.pp.normalize_per_cell(adata, counts_per_cell_after=1, copy=True)
     assert adata_copy.X.sum(axis=1).tolist() == [1.0, 1.0, 1.0]
     # now sparse
-    adata = AnnData(np.array([[1, 0], [3, 0], [5, 6]]))
-    adata_sparse = AnnData(sp.csr_matrix([[1, 0], [3, 0], [5, 6]]))
+    adata = AnnData(A.copy())
+    adata_sparse = AnnData(sp.csr_matrix(A.copy()))
     sc.pp.normalize_per_cell(adata)
     sc.pp.normalize_per_cell(adata_sparse)
     assert adata.X.sum(axis=1).tolist() == adata_sparse.X.sum(axis=1).A1.tolist()