Fix broken dtype checks in aesara.scalar.basic

aesara/scalar/basic.py

@@ -757,6 +757,8 @@ discrete_types = (bool,) + integer_types
 continuous_types = float_types + complex_types
 all_types = discrete_types + continuous_types
 
+discrete_dtypes = tuple(t.dtype for t in discrete_types)
+
 
 class _scalar_py_operators:
     # So that we can simplify checking code when we have a mixture of Scalar
@@ -1596,7 +1598,7 @@ class Switch(ScalarOp):
         first_part = switch(cond, gz, 0.0)
         second_part = switch(cond, 0.0, gz)
 
-        if outputs[0].type.dtype in discrete_types:
+        if outputs[0].type in discrete_types:
             first_part = 0.0
             second_part = 0.0
 
@@ -1991,7 +1993,7 @@ class TrueDiv(BinaryScalarOp):
     def impl(self, x, y):
         x = np.asarray(x)
         y = np.asarray(y)
-        if all(a.dtype in discrete_types for a in (x, y)):
+        if all(a.dtype in discrete_dtypes for a in (x, y)):
             return np.sctypeDict[config.floatX](float(x) / y)
         else:
             return x / y
@@ -2030,7 +2032,7 @@ class TrueDiv(BinaryScalarOp):
         # This is different from it not being connected
         # to the output; x/y is still a function of x
         # and y; it's just a step function.
-        if all(a.dtype in discrete_types for a in (x, y)):
+        if all(a.dtype in discrete_dtypes for a in (x, y)):
             return [x.zeros_like(), y.zeros_like()]
 
         first_part = gz / y
@@ -2255,7 +2257,7 @@ class Mod(BinaryScalarOp):
     def L_op(self, inputs, outputs, gout):
         (x, y) = inputs
         (gz,) = gout
-        if outputs[0].type.dtype in discrete_types:
+        if outputs[0].type in discrete_types:
             # The gradient does not flow in if the output is discrete
             return [
                 x.zeros_like(dtype=config.floatX),
@@ -2619,7 +2621,7 @@ class Sgn(UnaryScalarOp):
         (gz,) = gout
         rval = x.zeros_like()
 
-        if rval.type.dtype in discrete_types:
+        if rval.type in discrete_types:
             rval = rval.astype(config.floatX)
 
         return [rval]
@@ -2660,7 +2662,7 @@ class Ceil(UnaryScalarOp):
         (gz,) = gout
         rval = x.zeros_like()
 
-        if rval.type.dtype in discrete_types:
+        if rval.type in discrete_types:
             rval = rval.astype(config.floatX)
 
         return [rval]
@@ -2686,7 +2688,7 @@ class Floor(UnaryScalarOp):
         (gz,) = gout
         rval = x.zeros_like()
 
-        if rval.type.dtype in discrete_types:
+        if rval.type in discrete_types:
             rval = rval.astype(config.floatX)
 
         return [rval]
@@ -2740,7 +2742,7 @@ class RoundHalfToEven(UnaryScalarOp):
         (gz,) = gout
         rval = x.zeros_like()
 
-        if rval.type.dtype in discrete_types:
+        if rval.type in discrete_types:
             rval = rval.astype(config.floatX)
 
         return [rval]
@@ -2826,7 +2828,7 @@ class RoundHalfAwayFromZero(UnaryScalarOp):
         (gz,) = gout
         rval = x.zeros_like()
 
-        if rval.type.dtype in discrete_types:
+        if rval.type in discrete_types:
             rval = rval.astype(config.floatX)
 
         return [rval]