Python error when calling NumPy from class method with map

Question

The following code throws me the error:

Traceback (most recent call last):
  File \"\", line 25, in 
    sol = anna.main()
  File \"\", line 17, in         


        

Answer 1

It may not be your exact case, but if you want to take the square root of a large number AND you want to have control of the precision of the result, going either to math.sqrt() or x ** 0.5 would not be much useful because your result will end up being a float number, and for large enough inputs, this will hit the limitations of float numbers.

One possibility for the specific case of sqrt is to look into an integer-only implementation of the square root algorithm, e.g. flyingcircus.util.isqrt() (Disclaimer: I am the main author of flyingcircus).

Alternatively, for a more general solution, one may look into gmpy (or any other arbitrary precision library with Python bindings).

Answer 2

You can replace numpy by the built in function math.sqrt like this:

import math  

class anaconda():

    def __init__(self):
        self.mice = range(10000)

    def eat(self, food):
        calc = math.sqrt(food ** 5)
        return calc

    def main(self):

        sol = list(map(self.eat, self.mice))
        return sol


if __name__ == '__main__':
    anna = anaconda()
    sol = anna.main()
    print(len(sol))

I think that the problem of your code is that you are probably reaching a limit (not sure yet why it raises that confusing error) because 10000**5 is a reeeally big number. You can check this out by reducing your range(10000) to range(1000). You will notice that your code runs perfectly fine then:

import numpy as np  

class anaconda():

    def __init__(self):
        self.mice = range(1000)

    def eat(self, food):
        calc = np.sqrt((food ** 5))
        return calc

    def main(self):

        sol = list(map(self.eat, self.mice))
        print sol
        return sol


if __name__ == '__main__':
    anna = anaconda()
    sol = anna.main()
    print(len(sol))

This runs perfectly fine, just by reducing range(10000) to range(1000)

Answer 3

I'll try to add a precise answer to those that have already been given. numpy.sqrt has some limitations that math.sqrt doesn't have.

import math
import numpy  # version 1.13.3

print(math.sqrt(2 ** 64 - 1))
print(numpy.sqrt(2 ** 64 - 1))

print(math.sqrt(2 ** 64))
print(numpy.sqrt(2 ** 64))

returns (with Python 3.5) :

4294967296.0
4294967296.0
4294967296.0
Traceback (most recent call last):
  File "main.py", line 8, in <module>
    print(numpy.sqrt(2 ** 64))
AttributeError: 'int' object has no attribute 'sqrt'

In fact, 2 ** 64 is equal to 18,446,744,073,709,551,616 and, according to the standard of C data types (version C99), the long long unsigned integer type contains at least the range between 0 and 18,446,744,073,709,551,615 included.

The AttributeError occurs because numpy, seeing a type that it doesn't know how to handle (after conversion to C data type), defaults to calling the sqrt method on the object (but that doesn't exist). If we use floats instead of integers then everything will work using numpy:

import numpy  # version 1.13.3

print(numpy.sqrt(float(2 ** 64)))

returns:

4294967296.0

So instead of replacing numpy.sqrt by math.sqrt, you can alternatively replace calc = np.sqrt(food ** 5) by calc = np.sqrt(float(food ** 5)) in your code.

I hope this error will make more sense to you now.

Answer 4

Plain Python

Actually, you need neither numpy nor math because sqrt(x) is x**0.5. So:

sqrt(x**5) = x ** (5/2) = x ** 2.5

It means you could replace your code with:

class anaconda():
    def __init__(self):
        self.mice = range(10000)

    def eat(self, food):
        calc = food ** 2.5
        return calc

    def main(self):
        sol = list(map(self.eat, self.mice))
        return sol


if __name__ == '__main__':
    anna = anaconda()
    sol = anna.main()
    print(len(sol))

NumPy

If you want to use NumPy, you can enjoy the fact that you can work with arrays as if they were scalars:

import numpy as np

class anaconda():

    def __init__(self):
        self.mice = np.arange(10000)

    def eat(self, food):
        return food ** 2.5

    def main(self):
        return self.eat(self.mice)


if __name__ == '__main__':
    anna = anaconda()
    sol = anna.main()
    print(len(sol))

Short refactor

Removing all the unneeded object-oriented-with-weird-names, your code becomes:

import numpy as np
print(np.arange(10000) ** 2.5)

Answer 5

As others have noticed, this boils down to the fact that np.sqrt(7131 ** 5) works but np.sqrt(7132 ** 5) returns an error:

import numpy as np

print(np.sqrt(7131 ** 5))
print(np.sqrt(7132 ** 5))

# 4294138928.9
Traceback (most recent call last):
  File "main.py", line 4, in <module>
    print(np.sqrt(7132 ** 5))
AttributeError: 'int' object has no attribute 'sqrt'

Since np.sqrt docs don't mention any bounds on the argument, I'd consider this a numpy bug.