Implement specific make_node for sparse HStack/VStack with early static shape validation

pytensor/sparse/basic.py

@@ -1636,23 +1636,35 @@ class Stack(Op):
             raise ValueError("The output dtype must be specified.")
         self.dtype = dtype
 
-    def make_node(self, *mat):
-        if not mat:
+    def __str__(self):
+        return f"{self.__class__.__name__}({self.format},{self.dtype})"
+
+
+class HStack(Stack):
+    def make_node(self, *blocks):
+        if not blocks:
             raise ValueError("Cannot join an empty list of sparses.")
-        var = [as_sparse_variable(x) for x in mat]
+        input_blocks = [as_sparse_variable(block) for block in blocks]
 
-        for x in var:
+        for x in input_blocks:
             assert x.format in ("csr", "csc")
 
+        # Known rows numbers must be the same for all matrices.
+        static_n_rows = {
+            x.type.shape[0] for x in input_blocks if x.type.shape[0] is not None
+        }
+        if len(static_n_rows) > 1:
+            raise ValueError(
+                "All matrices must have the same number of rows; "
+                f"got row counts: {static_n_rows}."
+            )
+
         return Apply(
-            self, var, [SparseTensorType(dtype=self.dtype, format=self.format)()]
+            self,
+            input_blocks,
+            [SparseTensorType(dtype=self.dtype, format=self.format)()],
         )
 
-    def __str__(self):
-        return f"{self.__class__.__name__}({self.format},{self.dtype})"
-
-
-class HStack(Stack):
     def perform(self, node, block, outputs):
         (out,) = outputs
         for b in block:
@@ -1726,6 +1738,30 @@ def hstack(blocks, format=None, dtype=None):
 
 
 class VStack(Stack):
+    def make_node(self, *blocks):
+        if not blocks:
+            raise ValueError("Cannot join an empty list of sparses.")
+        input_blocks = [as_sparse_variable(block) for block in blocks]
+
+        for x in input_blocks:
+            assert x.format in ("csr", "csc")
+
+        # Known column numbers must be the same for all matrices.
+        static_n_cols = {
+            x.type.shape[1] for x in input_blocks if x.type.shape[1] is not None
+        }
+        if len(static_n_cols) > 1:
+            raise ValueError(
+                "All matrices must have the same number of columns; "
+                f"got column counts: {static_n_cols}."
+            )
+
+        return Apply(
+            self,
+            input_blocks,
+            [SparseTensorType(dtype=self.dtype, format=self.format)()],
+        )
+
     def perform(self, node, block, outputs):
         (out,) = outputs
         for b in block:

tests/link/numba/sparse/test_basic.py

@@ -409,8 +409,8 @@ def test_sparse_vstack(output_format, input_formats):
 
 
 def test_sparse_hstack_mismatched_rows_raises():
-    x = ps.matrix(name="x", shape=(3, 5), format="csr", dtype=config.floatX)
-    y = ps.matrix(name="y", shape=(4, 7), format="csr", dtype=config.floatX)
+    x = ps.matrix(name="x", shape=(None, 5), format="csr", dtype=config.floatX)
+    y = ps.matrix(name="y", shape=(None, 7), format="csr", dtype=config.floatX)
     z = ps.hstack([x, y], format="csr", dtype=config.floatX)
     fn = function([x, y], z, mode="NUMBA")
 
@@ -422,8 +422,8 @@ def test_sparse_hstack_mismatched_rows_raises():
 
 
 def test_sparse_vstack_mismatched_cols_raises():
-    x = ps.matrix(name="x", shape=(10, 3), format="csr", dtype=config.floatX)
-    y = ps.matrix(name="y", shape=(13, 4), format="csr", dtype=config.floatX)
+    x = ps.matrix(name="x", shape=(10, None), format="csr", dtype=config.floatX)
+    y = ps.matrix(name="y", shape=(13, None), format="csr", dtype=config.floatX)
     z = ps.vstack([x, y], format="csr", dtype=config.floatX)
     fn = function([x, y], z, mode="NUMBA")
 

tests/sparse/test_basic.py

@@ -1583,3 +1583,23 @@ def test_hstack_vstack():
             stacked_blocks = stack_function(blocks, dtype=to_dtype)
             expected_dtype = get_expected_dtype(blocks, to_dtype)
             assert stacked_blocks.dtype == expected_dtype
+
+
+def test_sparse_hstack_mismatched_static_rows_raises():
+    x = sparse.matrix(name="x", shape=(3, 5), format="csr", dtype=config.floatX)
+    y = sparse.matrix(name="y", shape=(4, 7), format="csr", dtype=config.floatX)
+
+    with pytest.raises(
+        ValueError, match="All matrices must have the same number of rows"
+    ):
+        sparse.hstack([x, y], format="csr", dtype=config.floatX)
+
+
+def test_sparse_vstack_mismatched_static_cols_raises():
+    x = sparse.matrix(name="x", shape=(10, 3), format="csr", dtype=config.floatX)
+    y = sparse.matrix(name="y", shape=(13, 4), format="csr", dtype=config.floatX)
+
+    with pytest.raises(
+        ValueError, match="All matrices must have the same number of columns"
+    ):
+        sparse.vstack([x, y], format="csr", dtype=config.floatX)