Implement JAX dot with sparse constants

Non-constant sparse inputs can't be handled because JAX does not allow Scipy sparse matrices as inputs. We could implement a BCOO type explicitly but this would be JAX exclusive, and the user would need to use it from the get go, meaning such graphs would not be compatible with other backends.

Author: ricardoV94

pytensor/link/jax/dispatch/__init__.py

@@ -12,5 +12,6 @@
 import pytensor.link.jax.dispatch.random
 import pytensor.link.jax.dispatch.elemwise
 import pytensor.link.jax.dispatch.scan
+import pytensor.link.jax.dispatch.sparse
 
 # isort: on

pytensor/link/jax/dispatch/sparse.py

@@ -0,0 +1,38 @@
+import jax.experimental.sparse as jsp
+from scipy.sparse import spmatrix
+
+from pytensor.graph.basic import Constant
+from pytensor.link.jax.dispatch import jax_funcify, jax_typify
+from pytensor.sparse.basic import Dot, StructuredDot
+from pytensor.sparse.type import SparseTensorType
+
+
+@jax_typify.register(spmatrix)
+def jax_typify_spmatrix(matrix, dtype=None, **kwargs):
+    # Note: This changes the type of the constants from CSR/CSC to BCOO
+    # We could add BCOO as a PyTensor type but this would only be useful for JAX graphs
+    # and it would break the premise of one graph -> multiple backends.
+    # The same situation happens with RandomGenerators...
+    return jsp.BCOO.from_scipy_sparse(matrix)
+
+
+@jax_funcify.register(Dot)
+@jax_funcify.register(StructuredDot)
+def jax_funcify_sparse_dot(op, node, **kwargs):
+    for input in node.inputs:
+        if isinstance(input.type, SparseTensorType) and not isinstance(input, Constant):
+            raise NotImplementedError(
+                "JAX sparse dot only implemented for constant sparse inputs"
+            )
+
+    if isinstance(node.outputs[0].type, SparseTensorType):
+        raise NotImplementedError("JAX sparse dot only implemented for dense outputs")
+
+    @jsp.sparsify
+    def sparse_dot(x, y):
+        out = x @ y
+        if isinstance(out, jsp.BCOO):
+            out = out.todense()
+        return out
+
+    return sparse_dot

tests/link/jax/test_sparse.py

@@ -0,0 +1,75 @@
+import numpy as np
+import pytest
+import scipy.sparse
+
+import pytensor.sparse as ps
+import pytensor.tensor as pt
+from pytensor import function
+from pytensor.graph import FunctionGraph
+from tests.link.jax.test_basic import compare_jax_and_py
+
+
+@pytest.mark.parametrize(
+    "op, x_type, y_type",
+    [
+        (ps.dot, pt.vector, ps.matrix),
+        (ps.dot, pt.matrix, ps.matrix),
+        (ps.dot, ps.matrix, pt.vector),
+        (ps.dot, ps.matrix, pt.matrix),
+        # structured_dot only allows matrix @ matrix
+        (ps.structured_dot, pt.matrix, ps.matrix),
+        (ps.structured_dot, ps.matrix, pt.matrix),
+    ],
+)
+def test_sparse_dot_constant_sparse(x_type, y_type, op):
+    inputs = []
+    test_values = []
+
+    if x_type is ps.matrix:
+        x_sp = scipy.sparse.random(5, 40, density=0.25, format="csr", dtype="float32")
+        x_pt = ps.as_sparse_variable(x_sp, name="x")
+    else:
+        x_pt = x_type("x", dtype="float32")
+        if x_pt.ndim == 1:
+            x_test = np.arange(40, dtype="float32")
+        else:
+            x_test = np.arange(5 * 40, dtype="float32").reshape(5, 40)
+        inputs.append(x_pt)
+        test_values.append(x_test)
+
+    if y_type is ps.matrix:
+        y_sp = scipy.sparse.random(40, 3, density=0.25, format="csc", dtype="float32")
+        y_pt = ps.as_sparse_variable(y_sp, name="y")
+    else:
+        y_pt = y_type("y", dtype="float32")
+        if y_pt.ndim == 1:
+            y_test = np.arange(40, dtype="float32")
+        else:
+            y_test = np.arange(40 * 3, dtype="float32").reshape(40, 3)
+        inputs.append(y_pt)
+        test_values.append(y_test)
+
+    dot_pt = op(x_pt, y_pt)
+    fgraph = FunctionGraph(inputs, [dot_pt])
+    compare_jax_and_py(fgraph, test_values)
+
+
+def test_sparse_dot_non_const_raises():
+    x_pt = pt.vector("x")
+
+    y_sp = scipy.sparse.random(40, 3, density=0.25, format="csc", dtype="float32")
+    y_pt = ps.as_sparse_variable(y_sp, name="y").type()
+
+    out = ps.dot(x_pt, y_pt)
+
+    msg = "JAX sparse dot only implemented for constant sparse inputs"
+
+    with pytest.raises(NotImplementedError, match=msg):
+        function([x_pt, y_pt], out, mode="JAX")
+
+    y_pt_shared = ps.shared(y_sp, name="y")
+
+    out = ps.dot(x_pt, y_pt_shared)
+
+    with pytest.raises(NotImplementedError, match=msg):
+        function([x_pt], out, mode="JAX")