isinstance() 和 np.issubdtype() 有什么区别？

Question

I have a dataframe df:

>>> df = pd.DataFrame({'values':[1.23, 3.12, 23.12]}, dtype=float)

Running the check:

>>> df.values.dtype
dtype('float64')

Now using isinstance method:

>>> isinstance(df.values.dtype, float)
False

Now using issubdtype method:

>>> np.issubdtype(df.values.dtype, float)
True

By reading the doc.

Syntax isinstance (object, classinfo)

object : An object instance.

I'm passing a ndarray object (Can I consider this as an object instance?)

classinfo : A class, type or a tuple containing classes, types or other tuples.

I'm passing the float which is a type, which is fine.

The question is: why is the isinstance not returning true in the above example? and what's the difference between isinstance and issubdtype?

Answer 1

why is the isinstance not returning true in the above example?

Because df.values.dtype returns a type , not an object of that type. Indeed, df.values.dtype returns dtype('float') , but that is not a float object . It is a dtype object.

The issubdtype takes a dtype, type, or a string and checks if the first type is the same or a subclass of the second type. We can inspect the method resolution order of the type, and see:

>>> df.values.dtype.type.__mro__
(<class 'numpy.float64'>, <class 'numpy.floating'>, <class 'numpy.inexact'>, <class 'numpy.number'>, <class 'numpy.generic'>, , <class 'object'>)

so the type is indeed a subclass of float . But it is not an instance of a float . An instance of a float is for example 0.0 , 3.14 , np.nan , etc. If we inspect the type of these objects, we get:

>>> type(0.0)
<class 'float'>
>>> type(3.14)
<class 'float'>
>>> type(np.nan)
<class 'float'>

Whereas the type of for the dtype('float') is dtype , and for float , it is just type :

>>> type(df.values.dtype)
<class 'numpy.dtype'>
>>> type(float)
<class 'type'>

We can thus check that the df.values.dtype is an instance of dtype and obtain:

>>> isinstance(df.values.dtype, np.dtype)
True

Answer 2

First, df.values.dtype is an instance of the numpy's class dtype , not of the Python's class float :

>>> df.values.dtype
dtype('float64')

>>> isinstance(df.values.dtype, np.dtype)
True

>>> isinstance(df.values.dtype, float)
False

Second, from the source code of the function np.issubdtype we see that when calling np.issubdtype(df.values.dtype, float) , the first argument df.values.dtype is converted to df.values.dtype.type (ie np.float64 ) and the second argument float to np.floating . Then what is returned is issubclass(np.float64, np.floating) , which is of course True .

So, in short, df.values.dtype is a subclass of np.floating , not an instance of float .