Implement dot for XTensorVariables (#1475)

Author: AllenDowney

pytensor/xtensor/__init__.py

@@ -2,6 +2,7 @@
 
 import pytensor.xtensor.rewriting
 from pytensor.xtensor import linalg
+from pytensor.xtensor.math import dot
 from pytensor.xtensor.shape import concat
 from pytensor.xtensor.type import (
     as_xtensor,

pytensor/xtensor/math.py

@@ -4,9 +4,11 @@
 
 import pytensor.scalar as ps
 from pytensor import config
+from pytensor.graph.basic import Apply
 from pytensor.scalar import ScalarOp
-from pytensor.scalar.basic import _cast_mapping
-from pytensor.xtensor.basic import as_xtensor
+from pytensor.scalar.basic import _cast_mapping, upcast
+from pytensor.xtensor.basic import XOp, as_xtensor
+from pytensor.xtensor.type import xtensor
 from pytensor.xtensor.vectorization import XElemwise
 
 
@@ -139,3 +141,110 @@ def cast(x, dtype):
 def softmax(x, dim=None):
     exp_x = exp(x)
     return exp_x / exp_x.sum(dim=dim)  # type: ignore
+
+
+class XDot(XOp):
+    """Matrix multiplication between two XTensorVariables.
+
+    This operation performs matrix multiplication between two tensors, automatically
+    aligning and contracting dimensions. The behavior matches xarray's dot operation.
+
+    Parameters
+    ----------
+    dims : tuple of str
+        The dimensions to contract over. If None, will contract over all matching dimensions.
+    """
+
+    __props__ = ("dims",)
+
+    def __init__(self, dims: Iterable[str]):
+        self.dims = dims
+        super().__init__()
+
+    def make_node(self, x, y):
+        x = as_xtensor(x)
+        y = as_xtensor(y)
+
+        x_shape_dict = dict(zip(x.type.dims, x.type.shape))
+        y_shape_dict = dict(zip(y.type.dims, y.type.shape))
+
+        # Check for dimension size mismatches (concrete only)
+        for dim in self.dims:
+            x_shape = x_shape_dict.get(dim, None)
+            y_shape = y_shape_dict.get(dim, None)
+            if (
+                isinstance(x_shape, int)
+                and isinstance(y_shape, int)
+                and x_shape != y_shape
+            ):
+                raise ValueError(f"Size of dim '{dim}' does not match")
+
+        # Determine output dimensions
+        shape_dict = {**x_shape_dict, **y_shape_dict}
+        out_dims = tuple(d for d in shape_dict if d not in self.dims)
+
+        # Determine output shape
+        out_shape = tuple(shape_dict[d] for d in out_dims)
+
+        # Determine output dtype
+        out_dtype = upcast(x.type.dtype, y.type.dtype)
+
+        out = xtensor(dtype=out_dtype, shape=out_shape, dims=out_dims)
+        return Apply(self, [x, y], [out])
+
+
+def dot(x, y, dim: str | Iterable[str] | EllipsisType | None = None):
+    """Matrix multiplication between two XTensorVariables.
+
+    This operation performs matrix multiplication between two tensors, automatically
+    aligning and contracting dimensions. The behavior matches xarray's dot operation.
+
+    Parameters
+    ----------
+    x : XTensorVariable
+        First input tensor
+    y : XTensorVariable
+        Second input tensor
+    dim : str, Iterable[Hashable], EllipsisType, or None, optional
+        The dimensions to contract over. If None, will contract over all matching dimensions.
+        If Ellipsis (...), will contract over all dimensions.
+
+    Returns
+    -------
+    XTensorVariable
+        The result of the matrix multiplication.
+
+    Examples
+    --------
+    >>> x = xtensor(dtype="float64", dims=("a", "b"), shape=(2, 3))
+    >>> y = xtensor(dtype="float64", dims=("b", "c"), shape=(3, 4))
+    >>> z = dot(x, y)  # Result has dimensions ("a", "c")
+    >>> z = dot(x, y, dim=...)  # Contract over all dimensions
+    """
+    x = as_xtensor(x)
+    y = as_xtensor(y)
+
+    x_dims = set(x.type.dims)
+    y_dims = set(y.type.dims)
+    intersection = x_dims & y_dims
+    union = x_dims | y_dims
+
+    # Canonicalize dims
+    if dim is None:
+        dim_set = intersection
+    elif dim is ...:
+        dim_set = union
+    elif isinstance(dim, str):
+        dim_set = {dim}
+    elif isinstance(dim, Iterable):
+        dim_set = set(dim)
+
+    # Validate provided dims
+    # Check if any dimension is not found in either input
+    for d in dim_set:
+        if d not in union:
+            raise ValueError(f"Dimension {d} not found in either input")
+
+    result = XDot(dims=tuple(dim_set))(x, y)
+
+    return result

pytensor/xtensor/rewriting/__init__.py

@@ -1,5 +1,6 @@
 import pytensor.xtensor.rewriting.basic
 import pytensor.xtensor.rewriting.indexing
+import pytensor.xtensor.rewriting.math
 import pytensor.xtensor.rewriting.reduction
 import pytensor.xtensor.rewriting.shape
 import pytensor.xtensor.rewriting.vectorization

pytensor/xtensor/rewriting/math.py

@@ -0,0 +1,47 @@
+from string import ascii_lowercase
+
+from pytensor.graph import node_rewriter
+from pytensor.tensor import einsum
+from pytensor.tensor.shape import specify_shape
+from pytensor.xtensor.basic import tensor_from_xtensor, xtensor_from_tensor
+from pytensor.xtensor.math import XDot
+from pytensor.xtensor.rewriting.utils import register_lower_xtensor
+
+
+@register_lower_xtensor
+@node_rewriter(tracks=[XDot])
+def lower_dot(fgraph, node):
+    """Rewrite XDot to tensor.dot.
+
+    This rewrite converts an XDot operation to a tensor-based dot operation,
+    handling dimension alignment and contraction.
+    """
+    [x, y] = node.inputs
+    [out] = node.outputs
+
+    # Convert inputs to tensors
+    x_tensor = tensor_from_xtensor(x)
+    y_tensor = tensor_from_xtensor(y)
+
+    # Collect all dimension names across inputs and output
+    all_dims = list(
+        dict.fromkeys(x.type.dims + y.type.dims + out.type.dims)
+    )  # preserve order
+    if len(all_dims) > len(ascii_lowercase):
+        raise ValueError("Too many dimensions to map to einsum subscripts")
+
+    dim_to_char = dict(zip(all_dims, ascii_lowercase))
+
+    # Build einsum string
+    x_subs = "".join(dim_to_char[d] for d in x.type.dims)
+    y_subs = "".join(dim_to_char[d] for d in y.type.dims)
+    out_subs = "".join(dim_to_char[d] for d in out.type.dims)
+    einsum_str = f"{x_subs},{y_subs}->{out_subs}"
+
+    # Perform the einsum operation
+    out_tensor = einsum(einsum_str, x_tensor, y_tensor)
+
+    # Reshape to match the output shape
+    out_tensor = specify_shape(out_tensor, out.type.shape)
+
+    return [xtensor_from_tensor(out_tensor, out.type.dims)]

pytensor/xtensor/type.py

@@ -649,6 +649,10 @@ def stack(self, dim, **dims):
     def unstack(self, dim, **dims):
         return px.shape.unstack(self, dim, **dims)
 
+    def dot(self, other, dim=None):
+        """Matrix multiplication with another XTensorVariable, contracting over matching or specified dims."""
+        return px.math.dot(self, other, dim=dim)
+
 
 class XTensorConstantSignature(TensorConstantSignature):
     pass

tests/xtensor/test_math.py

@@ -150,3 +150,167 @@ def test_cast():
     yc64 = x.astype("complex64")
     with pytest.raises(TypeError, match="Casting from complex to real is ambiguous"):
         yc64.astype("float64")
+
+
+def test_dot():
+    """Test basic dot product operations."""
+    # Test matrix-vector dot product (with multiple-letter dim names)
+    x = xtensor("x", dims=("aa", "bb"), shape=(2, 3))
+    y = xtensor("y", dims=("bb",), shape=(3,))
+    z = x.dot(y)
+    fn = xr_function([x, y], z)
+
+    x_test = DataArray(np.ones((2, 3)), dims=("aa", "bb"))
+    y_test = DataArray(np.ones(3), dims=("bb",))
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Test matrix-vector dot product with ellipsis
+    z = x.dot(y, dim=...)
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test, dim=...)
+    xr_assert_allclose(z_test, expected)
+
+    # Test matrix-matrix dot product
+    x = xtensor("x", dims=("a", "b"), shape=(2, 3))
+    y = xtensor("y", dims=("b", "c"), shape=(3, 4))
+    z = x.dot(y)
+    fn = xr_function([x, y], z)
+
+    x_test = DataArray(np.add.outer(np.arange(2.0), np.arange(3.0)), dims=("a", "b"))
+    y_test = DataArray(np.add.outer(np.arange(3.0), np.arange(4.0)), dims=("b", "c"))
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Test matrix-matrix dot product with string dim
+    z = x.dot(y, dim="b")
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test, dim="b")
+    xr_assert_allclose(z_test, expected)
+
+    # Test matrix-matrix dot product with list of dims
+    z = x.dot(y, dim=["b"])
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test, dim=["b"])
+    xr_assert_allclose(z_test, expected)
+
+    # Test matrix-matrix dot product with ellipsis
+    z = x.dot(y, dim=...)
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test, dim=...)
+    xr_assert_allclose(z_test, expected)
+
+    # Test a case where there are two dimensions to sum over
+    x = xtensor("x", dims=("a", "b", "c"), shape=(2, 3, 4))
+    y = xtensor("y", dims=("b", "c", "d"), shape=(3, 4, 5))
+    z = x.dot(y)
+    fn = xr_function([x, y], z)
+
+    x_test = DataArray(np.arange(24.0).reshape(2, 3, 4), dims=("a", "b", "c"))
+    y_test = DataArray(np.arange(60.0).reshape(3, 4, 5), dims=("b", "c", "d"))
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Same but with explicit dimensions
+    z = x.dot(y, dim=["b", "c"])
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test, dim=["b", "c"])
+    xr_assert_allclose(z_test, expected)
+
+    # Same but with ellipses
+    z = x.dot(y, dim=...)
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test, dim=...)
+    xr_assert_allclose(z_test, expected)
+
+    # Dot product with sum
+    x_test = DataArray(np.arange(24.0).reshape(2, 3, 4), dims=("a", "b", "c"))
+    y_test = DataArray(np.arange(60.0).reshape(3, 4, 5), dims=("b", "c", "d"))
+    expected = x_test.dot(y_test, dim=("a", "b", "c"))
+
+    x = xtensor("x", dims=("a", "b", "c"), shape=(2, 3, 4))
+    y = xtensor("y", dims=("b", "c", "d"), shape=(3, 4, 5))
+    z = x.dot(y, dim=("a", "b", "c"))
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Dot product with sum in the middle
+    x_test = DataArray(np.arange(120.0).reshape(2, 3, 4, 5), dims=("a", "b", "c", "d"))
+    y_test = DataArray(np.arange(360.0).reshape(3, 4, 5, 6), dims=("b", "c", "d", "e"))
+    expected = x_test.dot(y_test, dim=("b", "d"))
+    x = xtensor("x", dims=("a", "b", "c", "d"), shape=(2, 3, 4, 5))
+    y = xtensor("y", dims=("b", "c", "d", "e"), shape=(3, 4, 5, 6))
+    z = x.dot(y, dim=("b", "d"))
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Same but with first two dims
+    expected = x_test.dot(y_test, dim=["a", "b"])
+    z = x.dot(y, dim=["a", "b"])
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Same but with last two
+    expected = x_test.dot(y_test, dim=["d", "e"])
+    z = x.dot(y, dim=["d", "e"])
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Same but with every other dim
+    expected = x_test.dot(y_test, dim=["a", "c", "e"])
+    z = x.dot(y, dim=["a", "c", "e"])
+    fn = xr_function([x, y], z)
+    z_test = fn(x_test, y_test)
+    xr_assert_allclose(z_test, expected)
+
+    # Test symbolic shapes
+    x = xtensor("x", dims=("a", "b"), shape=(None, 3))  # First dimension is symbolic
+    y = xtensor("y", dims=("b", "c"), shape=(3, None))  # Second dimension is symbolic
+    z = x.dot(y)
+    fn = xr_function([x, y], z)
+    x_test = DataArray(np.ones((2, 3)), dims=("a", "b"))
+    y_test = DataArray(np.ones((3, 4)), dims=("b", "c"))
+    z_test = fn(x_test, y_test)
+    expected = x_test.dot(y_test)
+    xr_assert_allclose(z_test, expected)
+
+
+def test_dot_errors():
+    # No matching dimensions
+    x = xtensor("x", dims=("a", "b"), shape=(2, 3))
+    y = xtensor("y", dims=("b", "c"), shape=(3, 4))
+    with pytest.raises(ValueError, match="Dimension e not found in either input"):
+        x.dot(y, dim="e")
+
+    # Concrete dimension size mismatches
+    x = xtensor("x", dims=("a", "b"), shape=(2, 3))
+    y = xtensor("y", dims=("b", "c"), shape=(4, 5))
+    with pytest.raises(
+        ValueError,
+        match="Size of dim 'b' does not match",
+    ):
+        x.dot(y)
+
+    # Symbolic dimension size mismatches
+    x = xtensor("x", dims=("a", "b"), shape=(2, None))
+    y = xtensor("y", dims=("b", "c"), shape=(None, 5))
+    z = x.dot(y)
+    fn = xr_function([x, y], z)
+    x_test = DataArray(np.ones((2, 3)), dims=("a", "b"))
+    y_test = DataArray(np.ones((4, 5)), dims=("b", "c"))
+    # Doesn't fail until the rewrite
+    with pytest.raises(ValueError, match="not aligned"):
+        fn(x_test, y_test)