Create NumPy array from another array by specifying rows and columns - python

How can I create a NumPy array B which is a sub-array of a NumPy array A, by specifying which rows and columns (indicated by x and y respectively) are to be included?
For example:
A = numpy.array([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]])
x = [0, 2]
y = [1, 3, 4]
B = # Do something .....
Should give the output:
>>> B
array([[2, 4, 5], [12, 14, 15]])

The best way to do this is to use the ix_ function: see the answer by MSeifert for details.
Alternatively, you could use chain the indexing operations using x and y:
>>> A[x][:,y]
array([[ 2, 4, 5],
[12, 14, 15]])
First x is used to select the rows of A. Next, [:,y] picks out the columns of the subarray specified by the elements of y.
The chaining is symmetric in this case: you can also choose the columns first with A[:,y][x] if you wish.

You can use np.ix_ which allows to broadcast the integer index arrays:
>>> A = np.array([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]])
>>> x = [0, 2]
>>> y = [1, 3, 4]
>>> A[np.ix_(x, y)]
array([[ 2, 4, 5],
[12, 14, 15]])
From the documentation the ix_ function was designed so
[...] one can quickly construct index arrays that will index the cross product. a[np.ix_([1,3],[2,5])] returns the array [[a[1,2] a[1,5]], [a[3,2] a[3,5]]].

Here's a super verbose way to get what you want:
import numpy as np
a = np.array([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]])
x = [0, 2]
y = [1,3,4]
a2 = a.tolist()
a3 = [[l for k,l in enumerate(j) if k in y] for i,j in enumerate(a2) if i in x]
b = np.array(a3)
But please follow #ajcr answer:
import numpy as np
a = np.array([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]])
x = [0, 2]
y = [1,3,4]
a[x][:,y]

Related

How to slice tensorflow tensor differently for each row at once?

I have simple tensor
a = tf.constant([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]])
and want to slice it, but I need to do it differently for each of the rows. This slice operation is described by another tensor
b = tf.constant([[0, 1], [2, 4], [2, 5]])
It means that from the first row of tensor a I need elements from 0 to 1, from the second row from 2 to 4 and so on. So the final final result will be
[
[1],
[8, 9],
[13, 14, 15]
]
My first idea was to fill ranges between begin and end of a slice, but unfortunately, doing it with map_fn is not possible because result's rows have different lengths.
Does anyone know how to do such operation?
Basically we have two arrays to iterate. One with actual data, other with range to return.
Therefore, zip function can help iterate over elements from multiple arrays one by one.
import tensorflow as tf
a = tf.constant([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]])
b = tf.constant([[0, 1], [2, 4], [2, 5]])
# As you iterate, provided a and b have same length
# [1, 2, 3, 4, 5] sliced as [0:1]
# [6, 7, 8, 9, 10] sliced as [2:4]
# [11, 12, 13, 14, 15] sliced as [2:5]
[data.numpy().tolist()[start:end] for data, (start, end) in zip(a,b)]
Output:
[[1], [8, 9], [13, 14, 15]]
If the size of b is known at graph compile time, then you can slice each row separately.
import tensorflow as tf
a = tf.constant([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]])
b = tf.constant([[0, 1], [2, 4], [2, 5]])
r = []
for i in range(3):
bi = b[i]
r.append(a[i][bi[0]: bi[1]])
print(r)

How to get 2D array values based on 1D array values?

>>> image = np.arange(20).reshape((4, 5))
>>> image
array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14],
[15, 16, 17, 18, 19]])
>>> idx = [[2, 1], [2, 3], [3, 4]]
How do I get the values from the image array which coordinate is specified in idx? From the above code, I want to get the values 11 (image[2, 1]), 13 (image[2, 3]), and 19 (image[3, 4]). Thank you.
(if ya gonna use numpy, use numpy)
Make definitions:
>>> image = np.arange(20).reshape((4, 5))
>>> idx = np.array([[2, 1], [2, 3], [3, 4]]).T
Solution using fancy indexing capabilities of Numpy:
>>> image[tuple(idx)]
array([11, 13, 19])
You can use a list comprehension:
[image[x[0], x[1]] for x in idx]
>>> [11, 13, 19]

How to modify N columns of numpy array at the same time?

How to modify N columns of numpy array?? For example, I have a numpy array as follows:
P = array([[1, 2, 3, 8, 6],
[4, 5, 6, 4, 5]
[0,-2, 5, 3, 0]])
How do I change the elements of first, second and forth columns of P?
Use indexing:
Here is an example:
>>> P[:, [0, 1, 3]] += 10
>>>
>>> P
array([[11, 12, 3, 18, 6],
[14, 15, 6, 14, 5],
[10, 8, 5, 13, 0]])

How to set a value to elements in a column filtered by another array

I have an m X 3 matrix and an array of length m.
I want to do the following
a = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12], [13, 14, 15]])
b = np.array([1, 2, 1, 3, 3])
me = np.mean(a[np.where(b==1)][:, 0])
a[np.where(b==1)][:, 0] = me
The problem is that
a[np.where(b==1)][:, 0]
returns [1, 7] instead of [4, 4].
You are combining index arrays with slices:
[np.where(b==1)] is a index array, [:, 0] is a slice. The way you do it a copy is returned and therefore you set the new values on the copy. You should instead do:
a = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12], [13, 14, 15]])
b = np.array([1, 2, 1, 3, 3])
me = np.mean(a[np.where(b==1)][:, 0])
a[np.where(b==1), 0] = me
Also see https://docs.scipy.org/doc/numpy/user/basics.indexing.html for combining index arrays with slices.

Reverse diagonal on numpy python

let's say I have this:
(numpy array)
a=
[0 1 2 3],
[4 5 6 7],
[8 9 10 11]
to get [1,1] which is 5 its diagonal is zero; according to numpy, a.diagonal(0)= [0,5,10]. How do I get the reverse or the right to left diagonal [2,5,8] for [1,1]? Is this possible?
My original problem is an 8 by 8 (0:7).. I hope that helps
Get a new array each row reversed.
>>> import numpy as np
>>> a = np.array([
... [0, 1, 2, 3],
... [4, 5, 6, 7],
... [8, 9, 10, 11]
... ])
>>> a[:, ::-1]
array([[ 3, 2, 1, 0],
[ 7, 6, 5, 4],
[11, 10, 9, 8]])
>>> a[:, ::-1].diagonal(1)
array([2, 5, 8])
or using numpy.fliplr:
>>> np.fliplr(a).diagonal(1)
array([2, 5, 8])
Flip the array upside-down and use the same:
np.flipud(a).diagonal(0)[::-1]
Another way to achieve this is to use np.rot90
import numpy as np
a = np.array([[0, 1, 2, 3],
[4, 5, 6, 7],
[8, 9, 10, 11]])
my_diag = np.rot90(a).diagonal(-1)
Result:
>>> my_diag
array([2, 5, 8])
A number of answers so far. #Akavall is closest as you need to rotate or filip and transpose (equivilant operations). I haven't seen a response from the OP regarding expected behavior on the "long" part of the rectangle.
Generalized solution for a square matrix:
a = array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14],
[15, 16, 17, 18, 19],
[20, 21, 22, 23, 24]])
>>> [(i, np.rot90(a).diagonal(2*i-a.shape[0]+1)) for i in range(a.shape[0])]
[(0, array([0])),
(1, array([ 2, 6, 10])),
(2, array([ 4, 8, 12, 16, 20])),
(3, array([14, 18, 22])),
(4, array([24]))]
As a function:
def reverse_diag(arr, n):
idx = 2*n - arr.shape[0]+1
return np.rot90(arr).diagonal(idx)
original matrix can be made square with a[:np.min(a.shape),:np.min(a.shape)]
EDIT: OP indicated the array is square.... Final Answer is the above

Categories

Resources