How to slice a 2D Python Array? Fails with: "TypeError: list indices must be integers, not tuple"

I have a 2d array in the numpy module that looks like:

data = array([[1,2,3],
              [4,5,6],
              [7,8,9]])

I want to get a slice of this array that only includes certain columns of elemen开发者_如何学运维t. For example I may want columns 0 and 2:

data = [[1,3],
        [4,6],
        [7,9]]

What is the most Pythonic way to do this? (No for loops please)

I thought this would work:

newArray = data[:,[0,2]]

but it results in a:

TypeError: list indices must be integers, not tuple

The error say it explicitely : data is not a numpy array but a list of lists.

try to convert it to an numpy array first :

numpy.array(data)[:,[0,2]]

If you'd want to slice 2D list the following function may help

def get_2d_list_slice(self, matrix, start_row, end_row, start_col, end_col):
    return [row[start_col:end_col] for row in matrix[start_row:end_row]]

Actually, what you wrote should work just fine... What version of numpy are you using?

Just to verify, the following should work perfectly with any recent version of numpy:

import numpy as np
x = np.arange(9).reshape((3,3)) + 1
print x[:,[0,2]]

Which, for me, yields:

array([[1, 3],
       [4, 6],
       [7, 9]])

as it should...

THis may not be what you are looking for but this is would do. zip(*x)[whatever columns you might need]

Why it works on Numpy but not Python lists

Because with __getitem__ you can program you classes to do whatever you want with : and multiple arguments.

Numpy does this, but built-in lists do not.

More precisely:

class C(object):
    def __getitem__(self, k):
        return k

# Single argument is passed directly.
assert C()[0] == 0

# Multiple indices generate a tuple.
assert C()[0, 1] == (0, 1)

# Slice notation generates a slice object.
assert C()[1:2:3] == slice(1, 2, 3)

# If you omit any part of the slice notation, it becomes None.
assert C()[:] == slice(None, None, None)
assert C()[::] == slice(None, None, None)
assert C()[1::] == slice(1, None, None)
assert C()[:2:] == slice(None, 2, None)
assert C()[::3] == slice(None, None, 3)

# Tuple with a slice object:
assert C()[:, 1] == (slice(None, None, None), 1)

# Ellipsis class object.
assert C()[...] == Ellipsis

We can then open up slice objects as:

s = slice(1, 2, 3)
assert s.start == 1
assert s.stop == 2
assert s.step == 3

So that is why when you write:

[][1, 2]

Python says:

TypeError: list indices must be integers, not tuple

because you are trying to pass (1, 2) to the list's __getitem__, and built-in lists are not programmed to deal with tuple arguments, only integers.

Beware that numpy only accept regular array with the same size for each elements. you can somehow use : [a[i][0:2] for i in xrange(len(a))] it's pretty ugly but it works.

newArray = data[:,0:2]

or am I missing something?

The example in question begins with array, not with np.array, and array is not defined as an isolated prefix:

data = array([[1,2,3],
              [4,5,6],
              [7,8,9]])

data[:,[0,2]]

Error:

NameError: name 'array' is not defined

To reproduce the error, you need to drop that array frame (without np. in front, it does not have a definition anyway).

data = [[1,2,3],
        [4,5,6],
        [7,8,9]]

data[:,[0,2]]

Error:

TypeError: list indices must be integers or slices, not tuple

The user has probably used the inner list of the array for tests but asked the question with a copy from a np.array output. At least in 2021, the question is just plain wrong: it cannot be reproduced. And I doubt that the behaviour was different in 2010 (numpy is the basic package of python).

For completeness, as in the other answers:

data = np.array([[1,2,3],
                [4,5,6],
                [7,8,9]])

data[:,[0,2]]

Output:

array([[1, 3],
       [4, 6],
       [7, 9]])

You do not need a nested list to reproduce this. Slicing a one-dimensional list by two dimensions like with

[1,2][:, 0]

throws the same TypeError: list indices must be integers or slices, not tuple.