Why is that I'm able to successfully return a large np.float128 but when I try to create the same np.float128 in numpy I just get inf?

Question

I'm writing a function to take the exponent of a large number:

def exp_large_float(x):
    return np.exp(np.float128(np.array([x])))

result = exp_large_float(800.)
print(result)
>>> [2.72637457e+347]

print(result.dtype)
>>> float128

I'm testing that result is what I intend so I want to create a "ground truth" array and compare it to the output of exp_large_float:

ground_truth = np.float128(2.72637457e+347)
print(ground_truth)
>>> inf

Why is that I'm able to successfully return a np.float128 of 2.72637457e+347 but when I try to create this same thing in numpy I just get inf?

I'm using python version 2.7.15.

*Python* converts `2.72637457e+347` to `inf` before it gets passed to numpy. Python doesn't have 128 bit floating point values. — Warren Weckesser, Aug 30 '19 at 18:43
python can clearly handle a `np.float128(2.72637457e+347)` as I'm able to return it. How can I create this object? — NewNameStat, Aug 30 '19 at 18:45
You are returning a NumPy object. The Python interpreter doesn't handle `2.72637457e+347`. Try `print 2.72637457e+347` in the interactive shell. — Warren Weckesser, Aug 30 '19 at 18:47
@PaulPanzer Your comment worked! If you make it an answer I'll accept it. — NewNameStat, Aug 30 '19 at 18:48
@PaulPanzer, nice, I forgot that the NumPy objects will convert strings. You can also write `np.float128("2.72637457e+347") `. — Warren Weckesser, Aug 30 '19 at 18:49

score 4 · Accepted Answer · answered Aug 30 '19 at 18:57

As pointed out by @Warren Weckesser your code doesn't work because

2.72637457e+347

is a float literal in Python so it is interpreted as standard (presumably 64bit) float and becomes inf before it is passed on to the float128 factory.

You can avoid this by passing your argument as a string.

ground_truth = np.float128("2.72637457e+347")

Why is that I'm able to successfully return a large np.float128 but when I try to create the same np.float128 in numpy I just get inf?

1 Answers1