2

This question is related to the following post: Replacing Numpy elements if condition is met.

Suppose i have two, one-dimensional numpy arrays a and b, with 50 rows each.

I would like to create an array c of 50 rows, each of which will take the values 0-4 depending on whether a condition is met:

if a > 0 the value in the corresponding row of c should be 0
if a < 0 the value in the corresponding row of c should be 1
if a > 0 and b < 0 the value in the corresponding row of c should be 2
if b > 0 the value in the corresponding row of c should be 3

I suppose the broader question here is how can i assign specific values to an array when there are multiple conditions. I have tried variations from the post i referenced above but i have not been successful.

Any ideas of how i could achieve this, preferably without using a for-loop?

  • 1
    Some of your conditions contradict. For example, if `a > 0` then 0, and if `a > 0` and `b < 0` then 2. Is there priority? – user3483203 Mar 06 '18 at 16:27
  • The two conditions do not contradict in the sense that when `b<0` then `a>0 and b<0` is a separate condition in its own right. –  Mar 06 '18 at 17:18

3 Answers3

4

A straight forward solution would be to apply the assignments in sequence.

In [18]: a = np.random.choice([-1,1],size=(10,))
In [19]: b = np.random.choice([-1,1],size=(10,))
In [20]: a
Out[20]: array([-1,  1, -1, -1,  1, -1, -1,  1,  1, -1])
In [21]: b
Out[21]: array([-1,  1,  1,  1, -1,  1, -1,  1,  1,  1])

Start off with an array with the 'default' value:

In [22]: c = np.zeros_like(a)

Apply the second condition:

In [23]: c[a<0] = 1

The third requires a little care since it combines 2 tests. () matter here:

In [25]: c[(a>0)&(b<0)] = 2

And the last:

In [26]: c[b>0] = 3
In [27]: c
Out[27]: array([1, 3, 3, 3, 2, 3, 1, 3, 3, 3])

Looks like all of the initial 0s are overwritten.

With many elements in the arrays, and just a few tests, I wouldn't worry about speed. Focus on clarity and expressiveness, not compactness.

There is a 3 argument version of where that can choose between values or arrays. But I rarely use it, and don't see many questions about it either. 

In [28]: c = np.where(a>0, 0, 1)
In [29]: c
Out[29]: array([1, 0, 1, 1, 0, 1, 1, 0, 0, 1])
In [30]: c = np.where((a>0)&(b<0), 2, c)
In [31]: c
Out[31]: array([1, 0, 1, 1, 2, 1, 1, 0, 0, 1])
In [32]: c = np.where(b>0, 3, c)
In [33]: c
Out[33]: array([1, 3, 3, 3, 2, 3, 1, 3, 3, 3])

These wheres could be chained on one line.

c = np.where(b>0, 3, np.where((a>0)&(b<0), 2, np.where(a>0, 0, 1)))
hpaulj
  • 221,503
  • 14
  • 230
  • 353
  • The `where` chain is also sequential, correct? Or the ordering does not matter? –  Mar 06 '18 at 18:59
  • 1
    The inner most where is evaluated and the result is passed as argument to next. Standard python evaluation. – hpaulj Mar 06 '18 at 19:13
1

as @chrisz points out, you currently have overlapping conditions. This is how I'd go about using multiple if statements:

import numpy as np
a = np.random.random(50)*10 - 10
b = np.random.random(50)*10 - 10
c = [0*(a>0)*(b<0) + 1*(a<0) + 3*(a==0)*(b>0)]

The compare statements return 1 if true and 0 otherwise. By multiplying them and adding different statements you can make multiple if statements. However, this ONLY WORKS if the if statements don't overlap.

Nathan
  • 3,558
  • 1
  • 18
  • 38
1

A generalised solution to this problem is possible via np.select. You need only supply a list of conditions and choices:

np.random.seed(0)

a = np.random.randint(-10, 10, (5, 5))
b = np.random.randint(-10, 10, (5, 5))

conditions = [b > 0, (a > 0) & (b < 0), a < 0, a > 0]
choices = [3, 2, 1, 0]

res = np.select(conditions, choices)

array([[3, 3, 1, 3, 1],
       [3, 3, 3, 3, 3],
       [1, 2, 1, 1, 1],
       [0, 2, 3, 3, 1],
       [1, 2, 2, 2, 1]])
jpp
  • 159,742
  • 34
  • 281
  • 339