I have a numpy float array and an int array of the same length. I would like to concatenate them such that the output has the composite dtype (float, int). column_stacking them together just yields a float64 array:
import numpy
a = numpy.random.rand(5)
b = numpy.random.randint(0, 100, 5)
ab = numpy.column_stack([a, b])
print(ab.dtype)
float64
Any hints?
Create a 'blank' array:
In [391]: dt = np.dtype('f,i')
In [392]: arr = np.zeros(5, dtype=dt)
In [393]: arr
Out[393]:
array([(0., 0), (0., 0), (0., 0), (0., 0), (0., 0)],
dtype=[('f0', '<f4'), ('f1', '<i4')])
fill it:
In [394]: arr['f0']=np.random.rand(5)
In [396]: arr['f1']=np.random.randint(0,100,5)
In [397]: arr
Out[397]:
array([(0.40140057, 75), (0.93731374, 99), (0.6226782 , 48),
(0.01068745, 68), (0.19197434, 53)],
dtype=[('f0', '<f4'), ('f1', '<i4')])
There are recfunctions that can be used as well, but it's good to know (and understand) this basic approach.
np.zeros(...)from it, and assign the respective fields. Your link has more information on creating structured arrays. I don't think it mentionscolumn_stack.pd.DataFrame({'a':a, 'b':b})- should preserve dtypesnp.rec.fromarrays([a,b])if a recarray is ok