Seemingly inconsistent RuntimeWarning using numpy arange

Question

When my friend runs the following code, it gives a RuntimeWarning and returns "inf". However, if I run this same code, it executes without a warning and returns a value.

import numpy as np
total = 0
k = 10**6
arr = np.arange(k, 0, -1)
for i in arr:
    total += 1/(i**2)
print(total)

This is her output:

\Anaconda3\lib\site-packages\ipykernel_launcher.py:7: RuntimeWarning: divide by zero encountered in long_scalars
  import sys
inf

Replacing np.arange() with the built-in range() solves the problem, and she gets the same output as me.

My question is: what is causing this inconsistency between her output and mine? Additionally, why is she seeing a difference when using np.arange vs range, while for me the output is the same?

We are both using python 3.7.4 via Anaconda, and numpy 1.16.5, on 64-bit Windows 10.

first use print() to see values for arange() and range(). — furas
– furas, Commented May 27, 2020 at 14:38
it works correctly for both on LInux Mint (64-bit), Python 3.7.7 (64-bit) numpy 1.18.4, (without Anaconda). The same Python 2.7.17 (64-bit), numpy 1.16.6 Do you use 64-bit Python on both computers? — furas
– furas, Commented May 27, 2020 at 14:41
@furas printing appears to show the same values for both arange() and range(), just with arange() as a numpy array of course. — L.N.M
– L.N.M, Commented May 27, 2020 at 14:54

DrBwts · Accepted Answer · 2020-05-27 14:48:59Z

2

Its to do with what data type Numpy infers when you create arr. For me Numpy infered that I wanted np.int32 & that is just not big enough to hold (10**6)**2 so you end up with an inf value.

If you explcitly declare the data type as say np.float64 the problem resolves itself.

arr = np.arange(k, 0, -1, dtype=np.float64)

answered May 27, 2020 at 14:48

DrBwts

3,6876 gold badges48 silver badges70 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

L.N.M Over a year ago

Thanks, that does explain the issue on her end. But I don't understand why I am not seeing the same behavior. With the same version of numpy, why would numpy apparently infer different data types between my friend and I?

DrBwts Over a year ago

@L.N.M what value are you getting?

L.N.M Over a year ago

1.6449330668487263 @DrBwts

DrBwts Over a year ago

I must admit that's very odd. For W10 x64 machines the default should be int32 ie a c long int as opposed to a c long long int which your machine seems to be using.

Collectives™ on Stack Overflow

Seemingly inconsistent RuntimeWarning using numpy arange

1 Answer 1

4 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

4 Comments

Related