Fix imports

Author: asistradition

scself/_noise2self/common.py

@@ -147,7 +147,7 @@ def _noise_to_self_error(
 
     if (metric == 'mse' and is_csr(X)):
 
-        from ..sparse.graph import chunk_graph_mse
+        from scself.sparse.graph import chunk_graph_mse
 
         _n_row = X.shape[0]
 

scself/_noise2self/graph.py

@@ -6,7 +6,6 @@
 from sklearn.neighbors import NearestNeighbors
 
 from scself.sparse import is_csr
-from scself.sparse.graph import _shrink_sparse_graph_k
 from scself.utils import array_sum, dot
 
 
@@ -61,6 +60,8 @@ def _nn_slice(k, n):
     _smallest = keep == 'smallest'
 
     if neighbor_sparse:
+        from scself.sparse.graph import _shrink_sparse_graph_k
+
         _shrink_sparse_graph_k(
             neighbor_graph.data,
             neighbor_graph.indptr,
@@ -250,7 +251,7 @@ def _row_divide(arr, row_divide_vector):
         # Numba function for CSR
         # more memory efficient
         if is_csr(arr):
-            from ..sparse.math import _csr_row_divide
+            from scself.sparse.math import _csr_row_divide
 
             _csr_row_divide(
                 arr.data,

scself/scaling/truncscaler.py

@@ -16,7 +16,7 @@ def fit(self, X, y=None):
             # Use custom extractor to turn X into a CSC with no
             # indices array; RobustScaler makes an undesirabe
             # CSR->CSC conversion
-            from ..sparse.math import sparse_csr_extract_columns
+            from scself.sparse.math import sparse_csr_extract_columns
             super().fit(
                 sparse_csr_extract_columns(X, fake_csc_matrix=True),
                 y

scself/tests/test_mkl_hacks.py

@@ -14,7 +14,7 @@
 DATA = np.random.default_rng(100).random((100, 50)).astype(np.float32)
 
 
-@unittest.skipIf(MKL_SKIP, 'NO MKL')
+@unittest.skip
 class TestTruncatedSVDMKL(unittest.TestCase):
 
     @unittest.skipIf('CI' in os.environ, 'Skip for CI')

scself/utils/__init__.py

@@ -11,7 +11,10 @@
     coefficient_of_variation,
     variance
 )
-from ._pca import pca
+from ._pca import (
+    pca,
+    stratified_pca
+)
 from .shrink import shrink_array_to_zero_inplace
 from .cast_dtype_inplace import cast_to_float_inplace
 from .correlation import (

scself/utils/correlation.py

@@ -97,7 +97,7 @@ def correlation_clustering_and_umap(
         obs=pd.DataFrame(index=var_names) if var_names is not None else None
     )
 
-    # Special case handling when there are too many neighbors for
+    # Special case handling to silently handle when there are too many neighbors for
     # the provided data; comes up with submodules a lot
     if corr_dist_adata.shape[0] <= n_neighbors:
 

scself/utils/pairwise_loss.py

@@ -215,7 +215,7 @@ def mcv_mean_error(
 
     if sps.issparse(x):
 
-        from ..sparse.math import mcv_mean_error_sparse
+        from scself.sparse.math import mcv_mean_error_sparse
 
         return mcv_mean_error_sparse(
             x,

scself/utils/standardization.py

@@ -98,7 +98,7 @@ def _normalize(
         )
 
     elif is_csr(lref):
-        from ..sparse.math import sparse_normalize_total
+        from scself.sparse.math import sparse_normalize_total
         sparse_normalize_total(
             lref,
             size_factor=size_factor
@@ -133,7 +133,7 @@ def _normalize(
             scaler.scale_ = scale_factor
 
         if is_csr(lref):
-            from ..sparse.math import sparse_normalize_columns
+            from scself.sparse.math import sparse_normalize_columns
             sparse_normalize_columns(
                 lref,
                 scaler.scale_

scself/utils/sum.py

@@ -31,7 +31,7 @@ def array_sum(array, axis=None, squared=False):
     # Because the scipy sparse method here is
     # very memory inefficient
     else:
-        from ..sparse.math import sparse_sum
+        from scself.sparse.math import sparse_sum
 
         return sparse_sum(
             array,