XFAIL pytensor tests for uint64 data type

Author: Dhruvanshu-Joshi

pytensor/compile/function/__init__.py

@@ -312,7 +312,6 @@ def opt_log1p(node):
     else:
         # note: pfunc will also call orig_function -- orig_function is
         #      a choke point that all compilation must pass through
-
         fn = pfunc(
             params=inputs,
             outputs=outputs,

pytensor/compile/function/types.py

@@ -1758,7 +1758,6 @@ def orig_function(
             name=name,
             fgraph=fgraph,
         )
-        print(m)
         with config.change_flags(compute_test_value="off"):
             fn = m.create(defaults)
     finally:

pytensor/graph/op.py

@@ -291,7 +291,6 @@ def __call__(
 
         """
         node = self.make_node(*inputs, **kwargs)
-
         if name is not None:
             if len(node.outputs) == 1:
                 node.outputs[0].name = name

pytensor/link/jax/dispatch/nlinalg.py

@@ -104,56 +104,11 @@ def batched_dot(a, b):
     return batched_dot
 
 
-# @jax_funcify.register(Max)
-# @jax_funcify.register(Argmax)
-# def jax_funcify_MaxAndArgmax(op, **kwargs):
-#     axis = op.axis
-
-#     def maxandargmax(x, axis=axis):
-#         if axis is None:
-#             axes = tuple(range(x.ndim))
-#         else:
-#             axes = tuple(int(ax) for ax in axis)
-
-#         max_res = jnp.max(x, axis)
-
-#         # NumPy does not support multiple axes for argmax; this is a
-#         # work-around
-#         keep_axes = jnp.array(
-#             [i for i in range(x.ndim) if i not in axes], dtype="int64"
-#         )
-#         # Not-reduced axes in front
-#         transposed_x = jnp.transpose(
-#             x, jnp.concatenate((keep_axes, jnp.array(axes, dtype="int64")))
-#         )
-#         kept_shape = transposed_x.shape[: len(keep_axes)]
-#         reduced_shape = transposed_x.shape[len(keep_axes) :]
-
-#         # Numpy.prod returns 1.0 when arg is empty, so we cast it to int64
-#         # Otherwise reshape would complain citing float arg
-#         new_shape = (
-#             *kept_shape,
-#             jnp.prod(jnp.array(reduced_shape, dtype="int64"), dtype="int64"),
-#         )
-#         reshaped_x = transposed_x.reshape(new_shape)
-
-#         max_idx_res = jnp.argmax(reshaped_x, axis=-1).astype("int64")
-
-#         return max_res, max_idx_res
-
-#     return maxandargmax
-
-
 @jax_funcify.register(Max)
 def jax_funcify_Max(op, **kwargs):
     axis = op.axis
 
-    def max(x, axis=axis):
-        # if axis is None:
-        #     axes = tuple(range(x.ndim))
-        # else:
-        #     axes = tuple(int(ax) for ax in axis)
-
+    def max(x):
         max_res = jnp.max(x, axis)
 
         return max_res
@@ -165,7 +120,7 @@ def max(x, axis=axis):
 def jax_funcify_Argmax(op, **kwargs):
     axis = op.axis
 
-    def argmax(x, axis=axis):
+    def argmax(x):
         if axis is None:
             axes = tuple(range(x.ndim))
         else:

pytensor/tensor/math.py

@@ -107,6 +107,14 @@
     float64_atol = 1e-8
 
 
+def __getattr__(name):
+    if name == "MaxandArgmax":
+        raise AttributeError(
+            "The class `MaxandArgmax` has been deprecated. "
+            "Call `Max` and `Argmax` separately as an alternative."
+        )
+
+
 def _get_atol_rtol(a, b):
     tiny = ("float16",)
     narrow = ("float32", "complex64")
@@ -134,15 +142,6 @@ def _allclose(a, b, rtol=None, atol=None):
     return np.allclose(a, b, atol=atol_, rtol=rtol_)
 
 
-def __getattr__(name):
-    if name == "MaxandArgmax":
-        warnings.warn(
-            "The class `MaxandArgmax` has been deprecated. "
-            "Call `Max` and `Argmax` seperately as an alternative.",
-            FutureWarning,
-        )
-
-
 class Argmax(COp):
     """
     Calculate the argmax over a given axis or over all axes.
@@ -398,21 +397,21 @@ def max_and_argmax(a, axis=None, keepdims=False):
 
     """
     # Check axis and convert it to a Python list of integers.
-    # Axis will be used as an op param of MaxAndArgmax.
+    # Axis will be used as an op param of Max and Argmax.
     a = as_tensor_variable(a)
 
-    flag = False
+    is_axis_empty = False
     if axis == ():
-        flag = True
+        is_axis_empty = True
 
     axis = check_and_normalize_axes(a, axis)
 
-    if len(axis) == 0 and not flag:
+    if len(axis) == 0 and not is_axis_empty:
         axis = None
 
     out = Max(axis)(a)
 
-    if not flag:
+    if not is_axis_empty:
         argout = Argmax(axis)(a)
     else:
         argout = zeros_like(a, dtype="int64")
@@ -495,7 +494,8 @@ def grad(self, inp, grads):
         if g_max_disconnected:
             return [DisconnectedType()()]
 
-        if NoneConst.equals(axis):
+        # if NoneConst.equals(axis):
+        if axis is None:
             axis_ = list(range(x.ndim))
         else:
             axis_ = axis

pytensor/tensor/rewriting/uncanonicalize.py

@@ -57,7 +57,6 @@ def local_max_to_min(fgraph, node):
     """
     if node.op == neg and node.inputs[0].owner:
         max = node.inputs[0]
-        # print(max.owner.op.scalar_op)
         if (
             max.owner
             and isinstance(max.owner.op, CAReduce)

tests/link/numba/test_basic.py

@@ -256,8 +256,6 @@ def compare_numba_and_py(
     if assert_fn is None:
 
         def assert_fn(x, y):
-            print(x)
-            print(y)
             return np.testing.assert_allclose(x, y, rtol=1e-4) and compare_shape_dtype(
                 x, y
             )

tests/tensor/test_math.py

@@ -808,7 +808,11 @@ def test_basic_2(self, axis, np_axis):
         v_shape, i_shape = eval_outputs([vt.shape, it.shape])
         assert tuple(v_shape) == vt.type.shape
         assert tuple(i_shape) == it.type.shape
-        # Test valuesgi
+        # Test values
+        v, i = eval_outputs([vt, it])
+        assert i.dtype == "int64"
+        assert np.all(v == np_max)
+        assert np.all(i == np_argm)
 
     @pytest.mark.parametrize(
         "axis,np_axis",
@@ -1029,9 +1033,7 @@ def test_vectorize(self, core_axis, batch_axis):
         batch_x = tensor(shape=(3, 5, 5, 5, 5))
 
         # Test MaxAndArgmax
-        max_x, argmax_x = max_and_argmax(x, axis=core_axis)
-        max_node = max_x.owner
-        assert isinstance(max_node.op, Max)
+        argmax_x = argmax(x, axis=core_axis)
 
         arg_max_node = argmax_x.owner
         new_node = vectorize_node(arg_max_node, batch_x)
@@ -1394,6 +1396,7 @@ def _grad_list(self):
         # check_grad_max(data, eval_outputs(grad(max_and_argmax(n,
         # axis=1)[0], n)),axis=1)
 
+    @pytest.mark.xfail(reason="Fails due to #770")
     def test_uint(self):
         for dtype in ("uint8", "uint16", "uint32", "uint64"):
             itype = np.iinfo(dtype)
@@ -1420,7 +1423,7 @@ def test_bool(self):
 
 
 def test_MaxandArgmax_deprecated():
-    with pytest.warns(FutureWarning, match=".*deprecated.*"):
+    with pytest.raises(AttributeError):
         pytensor.tensor.math.MaxandArgmax
 
 

tests/tensor/utils.py

@@ -107,7 +107,6 @@ def inplace_func(
 def eval_outputs(outputs, ops=(), mode=None):
     f = inplace_func([], outputs, mode=mode)
     variables = f()
-
     if ops:
         assert any(isinstance(node.op, ops) for node in f.maker.fgraph.apply_nodes)
     if isinstance(variables, tuple | list) and len(variables) == 1: