Creating numpy.dtype from string without incosistencies

Question

Does numpy have a way to create dtypes from a string? numpy.dtype() is almost what I need, but it creates an inconsistency:

import numpy as np

dtype1 = np.dtype('float32')
dtype2 = np.float32

# This works (prints "same")
if dtype1 == dtype2:
    print("same")

# I want the code in the line below to produce a set with one item
# However, the resulting set is: {dtype('float32'), <class 'numpy.float32'>}
dtypes = {dtype1, dtype2}

# One way to fix the above is to write the following, but this is ugly
dtypes = {np.empty(1, dtype1).dtype, np.empty(1, dtype2).dtype}

Is there an elegant way to fix the above problem?

Thank you

Answer 1

While your two objects do the same thing when used as the dtype argument, they are not the same thing, or even the same kind of thing:

In [51]: np.dtype('float32')                                                                         
Out[51]: dtype('float32')
In [52]: np.float32                                                                                  
Out[52]: numpy.float32
In [53]: type(np.dtype('float32'))                                                                   
Out[53]: numpy.dtype
In [54]: type(np.float32)                                                                            
Out[54]: type

One is an instance of np.dtype, the other is a function.

np.empty(1, 'f').dtype is another string that produces the desired dtype, but clearly won't match in set.

Using the function:

In [59]: np.float32(1).dtype                                                                         
Out[59]: dtype('float32')