Why is it possible to use Python vanilla max() to numpy.ndarray

Question

ndarray is a datatype defined specifically in numpy, so I wonder why I can use the following:

import numpy as np
a = np.array([1,2,3,4])
max(a) --> 4

I assume that python converts this back to a list, but I don't understand how this is done automatically and how python "knows" HOW to convert it to a list?

In addition, this one doesn't work:

b = np.array([[1,2,3],[4,5,6]])
max(b)

so in this case, python obviously doesn't know HOW to convert this to a list. Why does numpy not automatically flatten the multidimensional array to a vector like this:

max(b.flatten(1))

thanks a lot in advance.

Answer 1

The Python function max does not expect just lists as an argument but instead accepts an iterable object as shown in the docs here:

max(iterable[, key])

It is not converting the numpy arrays to lists at all but is instead using the fact that numpy arrays are iterable (as are tuples, strings, etc).

Answer 2

The max function has the following arguments:

max(iterable[, key=func]) -> value
max(a, b, c, ...[, key=func]) -> value

In the numpy array case the numpy array is treated as an iterable. In the case of your b what will happen is that b gets iterated over. This results in:

>>> list(iter(b))
[array([1, 2, 3]), array([4, 5, 6])]

Now if you compare the two items with the default comparison function you'll get the same error:

>>> c = list(iter(b))
>>> cmp(c[0], c[1])

Numpy does not flatten the array as the iterator will loop, by default , over rows and not over elements. If you want more control over the iteration, you can use on of the array iterators such as:

the numpy.nditer function.

>>> max(np.nditer(b))
array(6)

or the .flat attribute:

>>> max(b.flat)
6