Unexpected output of python3 built-in max for array containing nan-CodePudding

When I try to find out the index of the largest element in a list by using the key keyword in the python3 built-in function, max, I find it doesn't necessarily return the index of the max element.
For example,

array = [float('nan'), 1.0, 3.0, 2.0, float('nan')]
max(range(len(array)), key=lambda i: array[i])

The above code yields 0 instead of 2 as expected.

Seemingly Related Questions

Find max values of 1D arrays that may contain NAN

CodePudding user response：

The code for max (and min) works roughly in the following way:

def my_max(sequence, key=None, default=None):
    max_value = None 
    max_item = None 
    for item in sequence:
        if key is not None:
            value = key(item)
        else:
            value = item 
        if max_item is None:
            max_item = item 
            max_value = value
        elif value > max_value:
            max_value = value  
            max_item = item 
    if max_item is None:
        if default is None:
            raise ValueError
        return default 
    return max_item

The actual source code (in C) can be found here.

The main takeaway here is that on the first iteration, for your given input, max_item is set to 0, and max_value is set to nan. One important thing to note is that the operations <, <=, >, >=, and == will always return False when a nan is involved, as defined by the IEEE 754 floating point standard (example source). This means that no greater element will ever be found after the first item, and hence 0 is returned.