FIX: Enforce sub-matrix shape to all-zero columns being dropped

CHANGES.md

@@ -1,5 +1,11 @@
 # Release history:
 
+## v1.1.1
+
+### Internal
+
+- FIX: Prevent Scipy from dropping columns that are all zero for sub-matrices
+
 ## v1.1.0
 
 Add new function to select top-n from blocks of a sparse matrix matmul.

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "sparse_dot_topn"
-version = "1.1.0"
+version = "1.1.1"
 description = "This package boosts a sparse matrix multiplication followed by selecting the top-n multiplication"
 readme = "README.md"
 requires-python = ">=3.8"

src/sparse_dot_topn/api.py

@@ -324,18 +324,15 @@ def zip_sp_matmul_topn(top_n: int, C_mats: list[csr_matrix]) -> csr_matrix:
         indptr.append(C.indptr)
         indices.append(C.indices)
 
+    ncols = np.asarray(ncols, int)
+    total_cols = ncols.sum()
     if not np.all(np.diff(_nrows) == 0):
         msg = "Each `C` in `C_mats` should have the same number of rows."
         raise ValueError(msg)
 
     return csr_matrix(
         _core.zip_sp_matmul_topn(
-            top_n=top_n,
-            Z_max_nnz=nrows * top_n,
-            nrows=nrows,
-            B_ncols=np.asarray(ncols, int),
-            data=data,
-            indptr=indptr,
-            indices=indices,
-        )
+            top_n=top_n, Z_max_nnz=nrows * top_n, nrows=nrows, B_ncols=ncols, data=data, indptr=indptr, indices=indices
+        ),
+        shape=(nrows, total_cols),
     )

tests/conftest.py

@@ -4,4 +4,4 @@
 
 @pytest.fixture(scope="package")
 def rng():
-    return np.random.Generator(np.random.PCG64DXSM(1483336117))
+    return np.random.Generator(np.random.PCG64DXSM(1481236117))

tests/test_api.py

@@ -224,7 +224,7 @@ def test_stack_zip_sp_matmul_topn(rng, dtype):
     C_ref = sp_matmul_topn(A, B.T, top_n=10, threshold=0.01, sort=True)
 
     # zipped C-matrix
-    As = [A[i*200:(i+1)*200] for i in range(5)]  # names-to-match split
+    As = [A[i * 200 : (i + 1) * 200] for i in range(5)]  # names-to-match split
     Bs = [B[:100], B[100:300], B[300:]]  # GT split
 
     Cs = [[sp_matmul_topn(Aj, Bi.T, top_n=10, threshold=0.01, sort=True) for Bi in Bs] for Aj in As]