Change np.seterr behavior inside a function only

I want to change the floating point error handling of numpy inside of a single function.

Typical example:

import numpy as np

def f():
    np.seterr(all='raise')
    # Do some other stuff

print(np.seterr())
# {'divide': 'warn', 'over': 'warn', 'under': 'ignore', 'invalid': 'warn'}

f()
print(np.seterr())
# {'divide': 'raise', 'over': 'raise', 'under': 'raise', 'invalid': 'raise'}

I would like the call to fto not change the output of np.seterr(), I have no idea how to do this.

Thanks for any help !

Answer

Use the numpy.errstate context manager to change floating-point error handling temporarily:

def f():
    with numpy.errstate(all='raise'):
        ...

Be aware that while this is a temporary change, it is still global across threads and coroutines. As of the time of this writing, NumPy doesn’t offer an option for thread-local or contextvars-based error settings.

As of NumPy 1.17, it is also possible to use numpy.errstate as a function decorator, with the same effect as wrapping the whole body in a with numpy.errstate(...): block:

@numpy.errstate(all='raise')
def f():
    ...