I have a following a function combination(). It returns a numpy array 'r'. In this 2D array has first column should be float. It returns it as string. It tried to convert column to float in following way.
RV = np.array([[0.23, 2.5 , 5. , 7.1],['a', 'b'],['a1', 'a2']])
def combination():
global r
r = np.array(np.meshgrid(*RV)).T.reshape(-1,len(RV))
return r
combination()
print(r)
r[:,0] = r[:,0].astype(float) #convert string column to float
print(r)
In [404]: RV = np.array([[0.23, 2.5 , 5. , 7.1],['a', 'b'],['a1', 'a2']])
In [405]: RV
Out[405]:
array([list([0.23, 2.5, 5.0, 7.1]), list(['a', 'b']), list(['a1', 'a2'])],
dtype=object)
RV is a object dtype array because the lists vary in size. It is essentially a list. In fact, give how you use it, you might as well leave it as a list.
In [406]: def combination(RV):
...: r = np.array(np.meshgrid(*RV)).T.reshape(-1,len(RV))
...: return r
In [407]: r = combination(RV)
In [408]: r
Out[408]:
array([['0.23', 'a', 'a1'],
['0.23', 'b', 'a1'],
['2.5', 'a', 'a1'],
['2.5', 'b', 'a1'],
['5.0', 'a', 'a1'],
['5.0', 'b', 'a1'],
['7.1', 'a', 'a1'],
['7.1', 'b', 'a1'],
['0.23', 'a', 'a2'],
['0.23', 'b', 'a2'],
['2.5', 'a', 'a2'],
['2.5', 'b', 'a2'],
['5.0', 'a', 'a2'],
['5.0', 'b', 'a2'],
['7.1', 'a', 'a2'],
['7.1', 'b', 'a2']], dtype='<U32')
r is a string dtype - all of it. You can convert a column, but you can't put the float values back into r (without them being converted back to strings).
In [409]: r[:,0].astype(float)
Out[409]:
array([0.23, 0.23, 2.5 , 2.5 , 5. , 5. , 7.1 , 7.1 , 0.23, 0.23, 2.5 ,
2.5 , 5. , 5. , 7.1 , 7.1 ])
meshgrid preserves the dtype when creating the list of arrays:
In [410]: np.meshgrid(*RV)
Out[410]:
[array([[[0.23, 0.23],
[2.5 , 2.5 ],
[5. , 5. ],
[7.1 , 7.1 ]],
[[0.23, 0.23],
[2.5 , 2.5 ],
[5. , 5. ],
[7.1 , 7.1 ]]]), array([[['a', 'a'],
['a', 'a'],
['a', 'a'],
['a', 'a']],
[['b', 'b'],
['b', 'b'],
['b', 'b'],
['b', 'b']]], dtype='<U1'), array([[['a1', 'a2'],
['a1', 'a2'],
['a1', 'a2'],
['a1', 'a2']],
[['a1', 'a2'],
['a1', 'a2'],
['a1', 'a2'],
['a1', 'a2']]], dtype='<U2')]
But when you wrap them in np.array it uses the common compatible dtype, string. You can individually reshape an element of that meshgrid list:
In [411]: _[0].ravel()
Out[411]:
array([0.23, 0.23, 2.5 , 2.5 , 5. , 5. , 7.1 , 7.1 , 0.23, 0.23, 2.5 ,
2.5 , 5. , 5. , 7.1 , 7.1 ])
Are you fully aware of consequences of making object dtype arrays?
By the way, look at this alternative RV:
In [416]: np.array([[0.23, 2.5],['a', 'b'],['a1', 'a2']])
Out[416]:
array([['0.23', '2.5'],
['a', 'b'],
['a1', 'a2']], dtype='<U32')
In [417]: np.array([[0.23, 2.5],['a', 'b'],['a1', 'a2']],object)
Out[417]:
array([[0.23, 2.5],
['a', 'b'],
['a1', 'a2']], dtype=object)
Reliably creating an object dtype array with a given shape is not a trivial task.
Numpy arrays are supposed to be homogeneous in nature. If you want the mix to data types you can use list specify data type as an object in numpy array.
Added dtype=object in both of your arrays.
import numpy as np
RV = np.array([[0.23, 2.5 , 5. , 7.1],['a', 'b'],['a1', 'a2']])
def combination():
global r
r = np.array(np.meshgrid(*RV), dtype=object).T.reshape(-1,len(RV))
return r
combination()
print(r)
Edit: For "RV" array you don't need to mention 'dtype=object' as it is already that. But, adding 'dtype=object' for 'r' np.array in the combination function will solve the need for any post-processing to convert them back to floats as suggested in other answers.
r