Removing columns and rows from ndarray Python 2.7

Question

I have implemented and algorithm to remove some columns and rows from an ndarray in Python 2.7, however I feel there should be a better way to do it. Probably I do not know how to do it well in Python, this is why I put the questions here. I have been searching but I have been not succesful finding similar questions and in the documentation (for example in slicing and indexing documentation from scipy)

Assume I have a ndarray with some rows and columns:

number_of_rows = 3
number_of_columns = 3
a = np.arange(number_of_rows*number_of_columns).reshape(number_of_rows,number_of_columns)
a

Which output is:

array([[0, 1, 2],
       [3, 4, 5],
       [6, 7, 8]])

Let's supposse I want to remove some columns and/or rows of the previous ndarray. In particular I want to remove column 0 and row 1, this is an output like:

array([[1, 2],
       [7, 8]])

To do that I follow the following steps, however they look too me not very elegant and I feel they should be a better implementation.

I select to columns and rows to remove, in this example:
```
rows_to_remove = [1]
columns_to_remove = [0]
```
Now I create a couple of list with the columns and rows to keep.
```
rows_to_keep = list(set(range(0,a.shape[0]))-set(rows_to_remove))
columns_to_keep = list(set(range(0,a.shape[1]))-set(columns_to_remove))
```
This step in Matlab will be simpler just by using ~ to slice the indexes of the matrix (in python ndarray). Is there a better way to do this?.
Then I select those columns and rows to keep:
```
a[rows_to_keep,:][:,columns_to_keep]
```

Output:

array([[1, 2],
       [7, 8]])

Please note that if you just write:

a[rows_to_keep,columns_to_keep]

which output is:

array([1, 8])

This is a little bit socking for me, that a[rows_to_keep,columns_to_keep] is different to a[rows_to_keep,:][:,columns_to_keep]. Is there a better way to cover those steps?

Thank you very much

For a slicing explanation: Slicing of a NumPy 2d array, or how do I extract an mxm submatrix from an nxn array (n>m)?(https://stackoverflow.com/questions/4257394/slicing-of-a-numpy-2d-array-or-how-do-i-extract-an-mxm-submatrix-from-an-nxn-ar?rq=1) — Rafael Valero, Jan 26 '18 at 10:08

I.Seck · Answer 1 · 2018-01-25T13:13:49.487

0

You could use the delete method to get this done: Using the array giving in your question as example. It would go like this:

number_of_rows = 3
number_of_columns = 3
a=np.arange(number_of_rows*number_of_columns).reshape(number_of_rows,number_of_columns)
b=np.delete(a,1,0)
b=np.delete(b,0,1)

And Voilà, b contains the output array, you want !!

edited Jan 25 '18 at 13:13

answered Jan 25 '18 at 13:03

I.Seck

11
5

Thanks. I should not do that for different reasons: a) the position is usefull to track back the data. b) later I want to use the whole matrix. Best wishes – Rafael Valero Jan 25 '18 at 13:20
You could put the positions of the rows and columns to delete in variables, before using the delete method. You could as well keep the original matrix ! – I.Seck Jan 25 '18 at 14:02
True, but it is a big matrix and from the point of view of memory it can be ineficient. – Rafael Valero Jan 26 '18 at 10:05

score 0 · Accepted Answer · answered Jan 29 '18 at 11:57

For Question 2)

Instead of :

a[rows_to_keep,:][:,columns_to_keep]

use:

a[np.ix_(rows_to_keep,columns_to_keep)].

This is called Advanced Indexing (see [Numpy documentation1 and Writting in sub-ndarray of a ndarray in the most pythonian way. Python 2 )

For Questions 1) I will use Question 2 previous solution: a) creation of a mask, there are more elegant ways to do it, for example see Create a boolean mask from an array, but for simplicity:

mask = np.zeros(a.shape,dtype=bool)
mask[rows_to_remove,:] = True
mask[:,columns_to_remove] = True

Now you can visualize:

a[~np.array(mask)]

Note now there is not need for Question 2) answer.

Summary:

mask = np.zeros(a.shape,dtype=bool)
mask[rows_to_remove,:] = True
mask[:,columns_to_remove] = True 
a[~np.array(mask)]

If wanted you can reshape:

a[~np.array(mask)].reshape(a.shape[0]-len(rows_to_remove),a.shape[1]-len(columns_to_remove))

Removing columns and rows from ndarray Python 2.7

2 Answers2