diff --git a/pandas/_testing/__init__.py b/pandas/_testing/__init__.py index 361998db8e..87d419e2db 100644 --- a/pandas/_testing/__init__.py +++ b/pandas/_testing/__init__.py @@ -111,6 +111,7 @@ ALL_FLOAT_DTYPES: list[Dtype] = [*FLOAT_NUMPY_DTYPES, *FLOAT_EA_DTYPES] COMPLEX_DTYPES: list[Dtype] = [complex, "complex64", "complex128"] STRING_DTYPES: list[Dtype] = [str, "str", "U"] +COMPLEX_FLOAT_DTYPES: list[Dtype] = [*COMPLEX_DTYPES, *FLOAT_NUMPY_DTYPES] DATETIME64_DTYPES: list[Dtype] = ["datetime64[ns]", "M8[ns]"] TIMEDELTA64_DTYPES: list[Dtype] = ["timedelta64[ns]", "m8[ns]"] diff --git a/pandas/conftest.py b/pandas/conftest.py index 7c35dfdde9..10134c90f8 100644 --- a/pandas/conftest.py +++ b/pandas/conftest.py @@ -1403,6 +1403,21 @@ def complex_dtype(request): return request.param +@pytest.fixture(params=tm.COMPLEX_FLOAT_DTYPES) +def complex_or_float_dtype(request): + """ + Parameterized fixture for complex and numpy float dtypes. + + * complex + * 'complex64' + * 'complex128' + * float + * 'float32' + * 'float64' + """ + return request.param + + @pytest.fixture(params=tm.SIGNED_INT_NUMPY_DTYPES) def any_signed_int_numpy_dtype(request): """ diff --git a/pandas/core/computation/expr.py b/pandas/core/computation/expr.py index b5861fbaeb..d642c37cea 100644 --- a/pandas/core/computation/expr.py +++ b/pandas/core/computation/expr.py @@ -31,7 +31,6 @@ from pandas.core.computation.ops import ( UNARY_OPS_SYMS, BinOp, Constant, - Div, FuncNode, Op, Term, @@ -370,7 +369,7 @@ class BaseExprVisitor(ast.NodeVisitor): "Add", "Sub", "Mult", - None, + "Div", "Pow", "FloorDiv", "Mod", @@ -533,9 +532,6 @@ class BaseExprVisitor(ast.NodeVisitor): left, right = self._maybe_downcast_constants(left, right) return self._maybe_evaluate_binop(op, op_class, left, right) - def visit_Div(self, node, **kwargs): - return lambda lhs, rhs: Div(lhs, rhs) - def visit_UnaryOp(self, node, **kwargs): op = self.visit(node.op) operand = self.visit(node.operand) diff --git a/pandas/core/computation/ops.py b/pandas/core/computation/ops.py index 95ac20ba39..ea8b1c0457 100644 --- a/pandas/core/computation/ops.py +++ b/pandas/core/computation/ops.py @@ -332,31 +332,6 @@ for d in (_cmp_ops_dict, _bool_ops_dict, _arith_ops_dict): _binary_ops_dict.update(d) -def _cast_inplace(terms, acceptable_dtypes, dtype) -> None: - """ - Cast an expression inplace. - - Parameters - ---------- - terms : Op - The expression that should cast. - acceptable_dtypes : list of acceptable numpy.dtype - Will not cast if term's dtype in this list. - dtype : str or numpy.dtype - The dtype to cast to. - """ - dt = np.dtype(dtype) - for term in terms: - if term.type in acceptable_dtypes: - continue - - try: - new_value = term.value.astype(dt) - except AttributeError: - new_value = dt.type(term.value) - term.update(new_value) - - def is_term(obj) -> bool: return isinstance(obj, Term) @@ -516,31 +491,6 @@ class BinOp(Op): def isnumeric(dtype) -> bool: return issubclass(np.dtype(dtype).type, np.number) - -class Div(BinOp): - """ - Div operator to special case casting. - - Parameters - ---------- - lhs, rhs : Term or Op - The Terms or Ops in the ``/`` expression. - """ - - def __init__(self, lhs, rhs) -> None: - super().__init__("/", lhs, rhs) - - if not isnumeric(lhs.return_type) or not isnumeric(rhs.return_type): - raise TypeError( - f"unsupported operand type(s) for {self.op}: " - f"'{lhs.return_type}' and '{rhs.return_type}'" - ) - - # do not upcast float32s to float64 un-necessarily - acceptable_dtypes = [np.float32, np.float64] - _cast_inplace(com.flatten(self), acceptable_dtypes, np.float64) - - UNARY_OPS_SYMS = ("+", "-", "~", "not") _unary_ops_funcs = (operator.pos, operator.neg, operator.invert, operator.invert) _unary_ops_dict = dict(zip(UNARY_OPS_SYMS, _unary_ops_funcs)) diff --git a/pandas/tests/computation/test_eval.py b/pandas/tests/computation/test_eval.py index 17630f14b0..e8fad6b8cb 100644 --- a/pandas/tests/computation/test_eval.py +++ b/pandas/tests/computation/test_eval.py @@ -747,16 +747,26 @@ class TestTypeCasting: @pytest.mark.parametrize("op", ["+", "-", "*", "**", "/"]) # maybe someday... numexpr has too many upcasting rules now # chain(*(np.core.sctypes[x] for x in ['uint', 'int', 'float'])) - @pytest.mark.parametrize("dt", [np.float32, np.float64]) @pytest.mark.parametrize("left_right", [("df", "3"), ("3", "df")]) - def test_binop_typecasting(self, engine, parser, op, dt, left_right): - df = DataFrame(np.random.default_rng(2).standard_normal((5, 3)), dtype=dt) + def test_binop_typecasting( + self, engine, parser, op, complex_or_float_dtype, left_right, request + ): + # GH#21374 + dtype = complex_or_float_dtype + df = DataFrame(np.random.default_rng(2).standard_normal((5, 3)), dtype=dtype) left, right = left_right s = f"{left} {op} {right}" res = pd.eval(s, engine=engine, parser=parser) - assert df.values.dtype == dt - assert res.values.dtype == dt - tm.assert_frame_equal(res, eval(s)) + if dtype == "complex64" and engine == "numexpr": + mark = pytest.mark.xfail( + reason="numexpr issue with complex that are upcast " + "to complex 128 " + "https://github.com/pydata/numexpr/issues/492" + ) + request.applymarker(mark) + assert df.values.dtype == dtype + assert res.values.dtype == dtype + tm.assert_frame_equal(res, eval(s), check_exact=False) # -------------------------------------