Allow overflow for numpy types

Question

I'm trying to get the "normal" overflow/underflow behavior of C-type languages in Python. To my surprise, a RuntimeWarning is raised when I'm trying to get this behavior. Example:

np.uint8(255) + np.uint8(1)
>>> RuntimeWarning: overflow encountered in ubyte_scalars

Is there any way to simulate the desired behavior, i.e., that 255+1 gives 0?

I tried the docs but cannot find this behavior documented.

user545424 · Accepted Answer · 2019-02-27 16:47:06Z

3

I believe numpy does give you the correct behavior.

In [1]: np.uint8(255) + np.uint8(1)
/usr/bin/ipython:1: RuntimeWarning: overflow encountered in ubyte_scalars
  #!/usr/bin/python2
Out[1]: 0

You can suppress the warning by running:

In [1]: np.seterr(over='ignore')
Out[1]: {'divide': 'warn', 'invalid': 'warn', 'over': 'warn', 'under': 'ignore'}

In [2]: np.uint8(255) + np.uint8(1)
Out[2]: 0

answered Feb 27, 2019 at 16:47

user545424

16.3k11 gold badges61 silver badges72 bronze badges

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

3 Comments

fabian789 Over a year ago

I see. So I guess these uint8 are by no means native? Is there any way to get fast uint8 semantics without resorting to a lower-level language?

user545424 Over a year ago

What do you mean by native? I think internally they are indeed represented as a uint8 same as in C, there is just some extra error checking going on.

user545424 Over a year ago

It appears that this check is only done for unsigned integers when adding scalars too since it would be too expensive to do for arrays. Have a look here: github.com/numpy/numpy/issues/8987.

Collectives™ on Stack Overflow

Allow overflow for numpy types

1 Answer 1

3 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related