I have a sample numpy array read from a file like this:
import numpy as np
data = np.array([(b'M', 119., 76.), (b'M', 114., 73.), (b'F', 124., 79.)],
dtype=[('gender', 'S1'), ('bps', '<f8'), ('bpd', '<f8')])
How to get the first column called gender?
I tried:
gender = data[:,0] # fails
gender = [i[0] for i in data] # pass
Is there easier way to get the first column without list comprehension?
Can we unpack 3 columns like g,b,p=*data?
You made a 1d structured array (with a compound dtype):
In [717]: data = np.array([(b'M', 119., 76.), (b'M', 114., 73.), (b'F', 124.,
...: 79.)],
...: dtype=[('gender', 'S1'), ('bps', '<f8'), ('bpd', '<f8')])
In [718]:
In [718]: data.shape
Out[718]: (3,)
In [719]: data.dtype.fields
Out[719]:
mappingproxy({'gender': (dtype('S1'), 0),
'bps': (dtype('float64'), 1),
'bpd': (dtype('float64'), 9)})
fields are accessed by name, not a column number:
In [720]: data['gender']
Out[720]: array([b'M', b'M', b'F'], dtype='|S1')
And element of this array is:
In [721]: data[0]
Out[721]: (b'M', 119., 76.)
In [722]: type(_)
Out[722]: numpy.void
We call that a record, but it does display (and index) as a tuple. Note that your creation expression uses a list of tuples.
One way of unpacking the array is to iterate over the field names:
In [734]: data.dtype.names
Out[734]: ('gender', 'bps', 'bpd')
In [735]: a,b,c = [data[name] for name in data.dtype.names]
In [736]: a
Out[736]: array([b'M', b'M', b'F'], dtype='|S1')
In [737]: b
Out[737]: array([119., 114., 124.])
In [738]: c
Out[738]: array([76., 73., 79.])
data.dtype.fields.index error solution and unpacking, SORRY, but Chris was the first to answer and crossed the milestone of 1k, so accepted his answer.First of all, your data doesn't have a column. It is just a 1d array.
data.shape
(3, )
So the slicing will raise IndexError as you passed 2d indices.
Try:
list(zip(*data))[0]
# (b'M', b'M', b'F')
This is not a ndarray , you have array(tuple,tuple,tuple)
data.shape
(3,)
Convert it to ndarray then you can slice by using the columns and index
np.array(data.tolist())[:,0]
array([b'M', b'M', b'F'], dtype='|S5')
More info :
np.array(data.tolist()).shape
(3, 3)
np.array at the beginning, I thought whatever is inside, we get numpy n-darry.
data[:,None] which might be simpler than tolist().