具有列访问权限的2D Numpy字符串数组

Question

New to Python and Numpy and MatPlotLib .

I am trying to create a 2D Numpy array from a CSV of various data types, but I will treat them all as strings. The killer is that I need to be able to access them with tuple indices, like: [:,5] to get the 5th column, or [5] to get the 5th row.

Is there any way to do this?

It seems that this is a limitation of Numpy due to the memory-access calculations:

dataSet = np.loadtxt(open("adult.data.csv", "rb"), delimiter=" ,")
print dataSet[:, 4] <---results in IndexError: Invalid Index

I have also tried loadfromgen , dtype = str and dtype = "a16" , as well as dtype = object . Nothing works. I can either load the data and it does not have column access, or I can't load the data at all.

Answer 1

Simulate you file from the comment line - replicated several time (ie one string per row of the file):

In [8]: txt = b" 39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K"
In [9]: txt = [txt for _ in range(5)]

In [10]: txt
Out[10]: 
[b' 39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K',
 b' 39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K',
 b' 39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K',
 b' 39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K',
 b' 39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K']

Load with genfromtxt , with delimiter. Let it choose the best dtype per column:

In [12]: A=np.genfromtxt(txt, delimiter=',',dtype=None)
In [13]: A
Out[13]: 
array([ (39, b' State-gov', 77516, b' Bachelors', 13, b' Never-married', b' Adm-clerical', b' Not-in-family', b' White', b' Male', 2174, 0, 40, b' United-States', b' <=50K'),
       (39, b' State-gov', 77516, b' Bachelors', 13, b' Never-married', b' Adm-clerical', b' Not-in-family', b' White', b' Male', 2174, 0, 40, b' United-States', b' <=50K'),...], 
      dtype=[('f0', '<i4'), ('f1', 'S10'), ('f2', '<i4'), ('f3', 'S10'), ('f4', '<i4'), ('f5', 'S14'), ('f6', 'S13'), ('f7', 'S14'), ('f8', 'S6'), ('f9', 'S5'), ('f10', '<i4'), ('f11', '<i4'), ('f12', '<i4'), ('f13', 'S14'), ('f14', 'S6')])

5 element array with a compound dtype

In [14]: A.shape
Out[14]: (5,)
In [15]: A.dtype
Out[15]: dtype([('f0', '<i4'), ('f1', 'S10'), ('f2', '<i4'),
    ('f3', 'S10'), ('f4', '<i4'), ....])

Access a 'column' with a field name (not column number)

In [16]: A['f4']
Out[16]: array([13, 13, 13, 13, 13])

Or load as dtype=str:

In [17]: A=np.genfromtxt(txt, delimiter=',',dtype=str)
In [18]: A
Out[18]: 
array([['39', ' State-gov', ' 77516', ' Bachelors', ' 13',
        ' Never-married', ' Adm-clerical', ' Not-in-family', ' White',
        ' Male', ' 2174', ' 0', ' 40', ' United-States', ' <=50K'],
        ...
        ' Male', ' 2174', ' 0', ' 40', ' United-States', ' <=50K']], 
      dtype='<U14')
In [19]: A.dtype
Out[19]: dtype('<U14')
In [20]: A.shape
Out[20]: (5, 15)
In [21]: A[:,4]
Out[21]: 
array([' 13', ' 13', ' 13', ' 13', ' 13'], 
      dtype='<U14')

Now it is 15 column 2d array that can be indexed with column number.

With the wrong delimiter, and it loads one column per row

In [24]: A=np.genfromtxt(txt, delimiter=' ,',dtype=str)
In [25]: A
Out[25]: 
array([ '39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K',
      ...], 
      dtype='<U127')
In [26]: A.shape
Out[26]: (5,)

A 1d array with a long string dtype.

A CSV file might loaded in various ways, some intentional, some not. You have to look at the results, and try to understand them before blindly trying to index columns.