Add full_like, ones_like, and zeros_like for XTensorVariable (#1514)

Author: AllenDowney

pytensor/xtensor/__init__.py

@@ -3,7 +3,7 @@
 import pytensor.xtensor.rewriting
 from pytensor.xtensor import linalg
 from pytensor.xtensor.math import dot
-from pytensor.xtensor.shape import concat
+from pytensor.xtensor.shape import concat, full_like, ones_like, zeros_like
 from pytensor.xtensor.type import (
     as_xtensor,
     xtensor,

pytensor/xtensor/shape.py

@@ -13,6 +13,7 @@
 from pytensor.tensor.type import integer_dtypes
 from pytensor.tensor.utils import get_static_shape_from_size_variables
 from pytensor.xtensor.basic import XOp
+from pytensor.xtensor.math import cast, second
 from pytensor.xtensor.type import as_xtensor, xtensor
 
 
@@ -498,3 +499,100 @@ def expand_dims(x, dim=None, create_index_for_new_dim=None, axis=None, **dim_kwa
         x = Transpose(dims=tuple(target_dims))(x)
 
     return x
+
+
+def full_like(x, fill_value, dtype=None):
+    """Create a new XTensorVariable with the same shape and dimensions, filled with a specified value.
+
+    Parameters
+    ----------
+    x : XTensorVariable
+        The tensor to fill.
+    fill_value : scalar or XTensorVariable
+        The value to fill the new tensor with.
+    dtype : str or np.dtype, optional
+        The data type of the new tensor. If None, uses the dtype of the input tensor.
+
+    Returns
+    -------
+    XTensorVariable
+        A new tensor with the same shape and dimensions as self, filled with fill_value.
+
+    Examples
+    --------
+    >>> from pytensor.xtensor import xtensor, full_like
+    >>> x = xtensor(dtype="float64", dims=("a", "b"), shape=(2, 3))
+    >>> y = full_like(x, 5.0)
+    >>> assert y.dims == ("a", "b")
+    >>> assert y.type.shape == (2, 3)
+    """
+    x = as_xtensor(x)
+    fill_value = as_xtensor(fill_value)
+
+    # Check that fill_value is a scalar (ndim=0)
+    if fill_value.type.ndim != 0:
+        raise ValueError(
+            f"fill_value must be a scalar, got ndim={fill_value.type.ndim}"
+        )
+
+    # Handle dtype conversion
+    if dtype is not None:
+        # If dtype is specified, cast the fill_value to that dtype
+        fill_value = cast(fill_value, dtype)
+    else:
+        # If dtype is None, cast the fill_value to the input tensor's dtype
+        # This matches xarray's behavior where it preserves the original dtype
+        fill_value = cast(fill_value, x.type.dtype)
+
+    # Use the xtensor second function
+    return second(x, fill_value)
+
+
+def ones_like(x, dtype=None):
+    """Create a new XTensorVariable with the same shape and dimensions, filled with ones.
+
+    Parameters
+    ----------
+    x : XTensorVariable
+        The tensor to fill.
+    dtype : str or np.dtype, optional
+        The data type of the new tensor. If None, uses the dtype of the input tensor.
+
+    Returns:
+    XTensorVariable
+        A new tensor with the same shape and dimensions as self, filled with ones.
+
+    Examples
+    --------
+    >>> from pytensor.xtensor import xtensor, full_like
+    >>> x = xtensor(dtype="float64", dims=("a", "b"), shape=(2, 3))
+    >>> y = ones_like(x)
+    >>> assert y.dims == ("a", "b")
+    >>> assert y.type.shape == (2, 3)
+    """
+    return full_like(x, 1.0, dtype=dtype)
+
+
+def zeros_like(x, dtype=None):
+    """Create a new XTensorVariable with the same shape and dimensions, filled with zeros.
+
+    Parameters
+    ----------
+    x : XTensorVariable
+        The tensor to fill.
+    dtype : str or np.dtype, optional
+        The data type of the new tensor. If None, uses the dtype of the input tensor.
+
+    Returns:
+    XTensorVariable
+        A new tensor with the same shape and dimensions as self, filled with zeros.
+
+    Examples
+    --------
+    >>> from pytensor.xtensor import xtensor, full_like
+    >>> x = xtensor(dtype="float64", dims=("a", "b"), shape=(2, 3))
+    >>> y = zeros_like(x)
+    >>> assert y.dims == ("a", "b")
+    >>> assert y.type.shape == (2, 3)
+    """
+    return full_like(x, 0.0, dtype=dtype)

tests/xtensor/test_shape.py

@@ -8,9 +8,10 @@
 from itertools import chain, combinations
 
 import numpy as np
+import xarray as xr
 from xarray import DataArray
-from xarray import concat as xr_concat
 
+import pytensor.xtensor as px
 from pytensor.tensor import scalar
 from pytensor.xtensor.shape import (
     concat,
@@ -226,7 +227,7 @@ def test_concat(dim):
     x3_test = xr_random_like(x3, rng)
 
     res = fn(x1_test, x2_test, x3_test)
-    expected_res = xr_concat([x1_test, x2_test, x3_test], dim=dim)
+    expected_res = xr.concat([x1_test, x2_test, x3_test], dim=dim)
     xr_assert_allclose(res, expected_res)
 
 
@@ -248,7 +249,7 @@ def test_concat_with_broadcast(dim):
     x3_test = xr_random_like(x3, rng)
     x4_test = xr_random_like(x4, rng)
     res = fn(x1_test, x2_test, x3_test, x4_test)
-    expected_res = xr_concat([x1_test, x2_test, x3_test, x4_test], dim=dim)
+    expected_res = xr.concat([x1_test, x2_test, x3_test, x4_test], dim=dim)
     xr_assert_allclose(res, expected_res)
 
 
@@ -263,7 +264,7 @@ def test_concat_scalar():
     x1_test = xr_random_like(x1)
     x2_test = xr_random_like(x2)
     res = fn(x1_test, x2_test)
-    expected_res = xr_concat([x1_test, x2_test], dim="new_dim")
+    expected_res = xr.concat([x1_test, x2_test], dim="new_dim")
     xr_assert_allclose(res, expected_res)
 
 
@@ -466,3 +467,148 @@ def test_expand_dims_errors():
     # Test with a numpy array as dim (not supported)
     with pytest.raises(TypeError, match="unhashable type"):
         y.expand_dims(np.array([1, 2]))
+
+
+def test_full_like():
+    """Test full_like function, comparing with xarray's full_like."""
+
+    # Basic functionality with scalar fill_value
+    x = xtensor("x", dims=("a", "b"), shape=(2, 3), dtype="float64")
+    x_test = xr_arange_like(x)
+
+    y1 = px.full_like(x, 5.0)
+    fn1 = xr_function([x], y1)
+    result1 = fn1(x_test)
+    expected1 = xr.full_like(x_test, 5.0)
+    xr_assert_allclose(result1, expected1, check_dtype=True)
+
+    # Other dtypes
+    x_3d = xtensor("x_3d", dims=("a", "b", "c"), shape=(2, 3, 4), dtype="float32")
+    x_3d_test = xr_arange_like(x_3d)
+
+    y7 = px.full_like(x_3d, -1.0)
+    fn7 = xr_function([x_3d], y7)
+    result7 = fn7(x_3d_test)
+    expected7 = xr.full_like(x_3d_test, -1.0)
+    xr_assert_allclose(result7, expected7, check_dtype=True)
+
+    # Integer dtype
+    y3 = px.full_like(x, 5.0, dtype="int32")
+    fn3 = xr_function([x], y3)
+    result3 = fn3(x_test)
+    expected3 = xr.full_like(x_test, 5.0, dtype="int32")
+    xr_assert_allclose(result3, expected3, check_dtype=True)
+
+    # Different fill_value types
+    y4 = px.full_like(x, np.array(3.14))
+    fn4 = xr_function([x], y4)
+    result4 = fn4(x_test)
+    expected4 = xr.full_like(x_test, 3.14)
+    xr_assert_allclose(result4, expected4, check_dtype=True)
+
+    # Integer input with float fill_value
+    x_int = xtensor("x_int", dims=("a", "b"), shape=(2, 3), dtype="int32")
+    x_int_test = DataArray(np.arange(6, dtype="int32").reshape(2, 3), dims=("a", "b"))
+
+    y5 = px.full_like(x_int, 2.5)
+    fn5 = xr_function([x_int], y5)
+    result5 = fn5(x_int_test)
+    expected5 = xr.full_like(x_int_test, 2.5)
+    xr_assert_allclose(result5, expected5, check_dtype=True)
+
+    # Symbolic shapes
+    x_sym = xtensor("x_sym", dims=("a", "b"), shape=(None, 3))
+    x_sym_test = DataArray(
+        np.arange(6, dtype=x_sym.type.dtype).reshape(2, 3), dims=("a", "b")
+    )
+
+    y6 = px.full_like(x_sym, 7.0)
+    fn6 = xr_function([x_sym], y6)
+    result6 = fn6(x_sym_test)
+    expected6 = xr.full_like(x_sym_test, 7.0)
+    xr_assert_allclose(result6, expected6, check_dtype=True)
+
+    # Boolean dtype
+    x_bool = xtensor("x_bool", dims=("a", "b"), shape=(2, 3), dtype="bool")
+    x_bool_test = DataArray(
+        np.array([[True, False, True], [False, True, False]]), dims=("a", "b")
+    )
+
+    y8 = px.full_like(x_bool, True)
+    fn8 = xr_function([x_bool], y8)
+    result8 = fn8(x_bool_test)
+    expected8 = xr.full_like(x_bool_test, True)
+    xr_assert_allclose(result8, expected8, check_dtype=True)
+
+    # Complex dtype
+    x_complex = xtensor("x_complex", dims=("a", "b"), shape=(2, 3), dtype="complex64")
+    x_complex_test = DataArray(
+        np.arange(6, dtype="complex64").reshape(2, 3), dims=("a", "b")
+    )
+
+    y9 = px.full_like(x_complex, 1 + 2j)
+    fn9 = xr_function([x_complex], y9)
+    result9 = fn9(x_complex_test)
+    expected9 = xr.full_like(x_complex_test, 1 + 2j)
+    xr_assert_allclose(result9, expected9, check_dtype=True)
+
+    # Symbolic fill value
+    x_sym_fill = xtensor("x_sym_fill", dims=("a", "b"), shape=(2, 3), dtype="float64")
+    fill_val = xtensor("fill_val", dims=(), shape=(), dtype="float64")
+    x_sym_fill_test = xr_arange_like(x_sym_fill)
+    fill_val_test = DataArray(3.14, dims=())
+
+    y10 = px.full_like(x_sym_fill, fill_val)
+    fn10 = xr_function([x_sym_fill, fill_val], y10)
+    result10 = fn10(x_sym_fill_test, fill_val_test)
+    expected10 = xr.full_like(x_sym_fill_test, 3.14)
+    xr_assert_allclose(result10, expected10, check_dtype=True)
+
+    # Test dtype conversion to bool when neither input nor fill_value are bool
+    x_float = xtensor("x_float", dims=("a", "b"), shape=(2, 3), dtype="float64")
+    x_float_test = xr_arange_like(x_float)
+
+    y11 = px.full_like(x_float, 5.0, dtype="bool")
+    fn11 = xr_function([x_float], y11)
+    result11 = fn11(x_float_test)
+    expected11 = xr.full_like(x_float_test, 5.0, dtype="bool")
+    xr_assert_allclose(result11, expected11, check_dtype=True)
+
+    # Verify the result is actually boolean
+    assert result11.dtype == "bool"
+    assert expected11.dtype == "bool"
+
+
+def test_full_like_errors():
+    """Test full_like function errors."""
+    x = xtensor("x", dims=("a", "b"), shape=(2, 3), dtype="float64")
+    x_test = xr_arange_like(x)
+
+    with pytest.raises(ValueError, match="fill_value must be a scalar"):
+        px.full_like(x, x_test)
+
+
+def test_ones_like():
+    """Test ones_like function, comparing with xarray's ones_like."""
+    x = xtensor("x", dims=("a", "b"), shape=(2, 3), dtype="float64")
+    x_test = xr_arange_like(x)
+
+    y1 = px.ones_like(x)
+    fn1 = xr_function([x], y1)
+    result1 = fn1(x_test)
+    expected1 = xr.ones_like(x_test)
+    xr_assert_allclose(result1, expected1)
+    assert result1.dtype == expected1.dtype
+
+
+def test_zeros_like():
+    """Test zeros_like function, comparing with xarray's zeros_like."""
+    x = xtensor("x", dims=("a", "b"), shape=(2, 3), dtype="float64")
+    x_test = xr_arange_like(x)
+
+    y1 = px.zeros_like(x)
+    fn1 = xr_function([x], y1)
+    result1 = fn1(x_test)
+    expected1 = xr.zeros_like(x_test)
+    xr_assert_allclose(result1, expected1)
+    assert result1.dtype == expected1.dtype

tests/xtensor/util.py

@@ -37,11 +37,30 @@ def xfn(*xr_inputs):
     return xfn
 
 
-def xr_assert_allclose(x, y, *args, **kwargs):
-    # Assert that two xarray DataArrays are close, ignoring coordinates
+def xr_assert_allclose(x, y, check_dtype=False, *args, **kwargs):
+    """Assert that two xarray DataArrays are close, ignoring coordinates.
+
+    Mostly a wrapper around xarray.testing.assert_allclose,
+    but with the option to check the dtype.
+
+    Parameters
+    ----------
+    x : xarray.DataArray
+        The first xarray DataArray to compare.
+    y : xarray.DataArray
+        The second xarray DataArray to compare.
+    check_dtype : bool, optional
+        If True, check that the dtype of the two DataArrays is the same.
+    *args :
+        Additional arguments to pass to xarray.testing.assert_allclose.
+    **kwargs :
+        Additional keyword arguments to pass to xarray.testing.assert_allclose.
+    """
     x = x.drop_vars(x.coords)
     y = y.drop_vars(y.coords)
     assert_allclose(x, y, *args, **kwargs)
+    if check_dtype:
+        assert x.dtype == y.dtype
 
 
 def xr_arange_like(x):