Okey I don't even know where to ask this or what I am exactly asking, but is basically what the title says: if the argument of numpy.min is a function, it doesn't matter if its defined by the user or not, the output is such function.

Let me put an example:

import numpy as np

def get_third_value(x): 
   return x[2]

xx = np.arange(10)

print(np.min(get_third_value)) # output:  <function the_third at 0x7ffb58c2b240>
print(np.min(max)) # output: <built-in function max>

# you can even use them: 
print(np.min(max)(xx)) # output : 9
print(np.min(get_third_value)(xx)) #  output: 2

Why this happens? Is this the intended behavior? If so, why?

I have checked other numpy functions and the same happens in some of them

Okey I don't even know where to ask this or what I am exactly asking, but is basically what the title says: if the argument of numpy.min is a function, it doesn't matter if its defined by the user or not, the output is such function.

Let me put an example:

import numpy as np

def get_third_value(x): 
   return x[2]

xx = np.arange(10)

print(np.min(get_third_value)) # output:  <function the_third at 0x7ffb58c2b240>
print(np.min(max)) # output: <built-in function max>

# you can even use them: 
print(np.min(max)(xx)) # output : 9
print(np.min(get_third_value)(xx)) #  output: 2

Why this happens? Is this the intended behavior? If so, why?

I have checked other numpy functions and the same happens in some of them

• python
• numpy

• 4 What were you expecting print(np.min(max)) to do? – khelwood Commented Mar 27 at 13:10
• Really nothing, it was a mistake and I was surprised that it didn't raise any error. – Alimanco Commented Mar 27 at 15:12
• This question might help. np.array(some_non_array_like_value) appears to return a 0-dimensional "array" that wraps the value, and for some reason when np.min is called on it it returns a value equivalent to the original one: np.min(True) == True is true, but np.min(True) is True is not, while np.min(max) is max is true. – bfontaine Commented Mar 28 at 9:31

1 Answer 1

In my current ipython session I have a foo function. In python functions are 'first class objects', that is, they can be passed around like other objects such as numbers and strings.

In [361]: foo
Out[361]: <function __main__.foo(a, b_list)>

In [362]: np.min(foo)
Out[362]: <function __main__.foo(a, b_list)>

np.min, like many numpy functions, turns its non-array arguments into an array first:

In [363]: np.array(foo)
Out[363]: array(<function foo at 0x000001595F1CBF60>, dtype=object)

In this case, and possibly many more, np.min just returns the only element of such a length 1 (0d) array.

Compare this case of a python float:

In [364]: np.min(1.23)
Out[364]: np.float64(1.23)

In [365]: np.array(1.23)
Out[365]: array(1.23)

In [366]: _.dtype
Out[366]: dtype('float64')

That float was made into an array with np.float64 dtype.

Look at what happens when we give it two functions:

In [367]: np.min([foo,foo])
TypeError: '<=' not supported between instances of 'function' and 'function'

Now it says it can't compare them. Now it's trying to do some sort of <=> comparison between the elements of this list. As an experiment, define a custom class, with (or without) at least one of the comparison methods, and see what np.min does.

Another exception to returning a single item:

In [368]: np.min('astring')
UFuncTypeError: ufunc 'minimum' did not contain a loop with signature matching types (dtype('<U7'), dtype('<U7')) -> None


In [369]: np.array('astring')
Out[369]: array('astring', dtype='<U7')

The string was turned into a numpy string dtype.

Functions np.min have a catalog of signatures, telling it what can be worked on. (I first wrote that np.min is a ufunc. That's not true. np.minimum is a ufunc. I have an idea of how to look up the signatures of ufunc. I don't know about this type.)

If the string is first made into an object dtype array, it behaves just like your function case:

In [370]: np.min(np.array('string',object))
Out[370]: 'string'

In general when trying to understand the bahavior with non-array inputs, first look at what np.array(...) does with that. Without digging further into the compiled code it will be hard to expain exactly what/why this is happending.

python min

Compare what the python min does. It expacts an iterable, like a list:

In [372]: min(foo)
TypeError: 'function' object is not iterable


In [373]: min([foo])
Out[373]: <function __main__.foo(a, b_list)>

In [374]: min([foo,foo])
TypeError: '<' not supported between instances of 'function' and 'function'