Closed
Description
Raised by this stackoverflow question. The following code , run on master...
x_actual = 49997
x3_actual = x_actual ** 3
for dtype in [np.int64, np.uint64, np.float64]:
x = {}
x['scalar'] = dtype(x_actual)
x[ '0d'] = np.array(x_actual, dtype=dtype)
x[ '1d'] = np.array([x_actual], dtype=dtype)
x[ '2d'] = np.array([[x_actual]], dtype=dtype)
x3 = {k: x**3 for k, x in x.items()}
assert all(v == x3_actual for v in x3.values()) #ok
print("input type of {}:".format(np.dtype(dtype)))
for k, v in x3.items():
print("{:>8s}: {}".format(k, v.dtype))
print()Gives the alarming output of:
input of int64
scalar: int64
0d: int64
1d: int64
2d: int64
input of uint64
scalar: float64
0d: float64
1d: uint64 # What?
2d: uint64 # What?
input of float64
scalar: float64
0d: float64
1d: float64
2d: float64
The return dtype of power should not depend on the shape of the input array