16
votes

I have come across some python code with slice notation that I am having trouble figuring out. It looks like slice notation but uses a comma and a list:

list[:, [1, 2, 3]]

Is this syntax valid? If so what does it do?

edit looks like it is a 2D numpy array

2
Are you certain list is a list? It could be an instance of a class that overrides __getitem__ in an unusual way. - Kevin
Have you tried type(list)? Could be a numpy array or pandas dataframe. - mattexx
@mattexx, I dont have the whole code, what does this do in numpy and pandas? - Stephan
Numpy arrays and pandas dataframes use 2-d arrays, much like matlab and r. The first argument typically represents a slice of rows, while the second represents a slice of columns. So list[:,:3] would be all rows, first three columns. I'm not sure if either supports the exact notation you have in your question. - mattexx
@mattexx: just so the OP isn't confused, it might be useful to emphasize that numpy arrays aren't fundamentally 2-D objects in the same way that some other datastructures are. They're n-D. - DSM

2 Answers

19
votes

Assuming that the object is really a numpy array, this is known as advanced indexing, and picks out the specified columns:

>>> import numpy as np
>>> a = np.arange(12).reshape(3,4)
>>> a
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
>>> a[:, [1,2,3]]
array([[ 1,  2,  3],
       [ 5,  6,  7],
       [ 9, 10, 11]])
>>> a[:, [1,3]]
array([[ 1,  3],
       [ 5,  7],
       [ 9, 11]])

Note that this won't work with the standard Python list:

>>> a.tolist()
[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9, 10, 11]]
>>> a.tolist()[:,[1,2,3]]
Traceback (most recent call last):
  File "<ipython-input-17-7d77de02047a>", line 1, in <module>
    a.tolist()[:,[1,2,3]]
TypeError: list indices must be integers, not tuple
3
votes

It generates a complex value and passes it to __*item__():

>>> class Foo(object):
...   def __getitem__(self, val):
...     print val
... 
>>> Foo()[:, [1, 2, 3]]
(slice(None, None, None), [1, 2, 3])

What it actually performs depends on the type being indexed.