David - 4 months ago 8x

Python Question

Imagine I have something like this:

`import numpy as np`

arra = np.arange(16).reshape(2, 2, 4)

which gives

`array([[[0, 1, 2, 3],`

[4, 5, 6, 7]],

[[8, 9, 10, 11],

[12, 13, 14, 15]]])

and I want to make a loop that runs along specific subarrays (in this case, I want to run along each 'column' of each 'matrix') and sum the result (that is, selecting 0 and 4 and summing them (4), selecting 1 and 5 and summing them (6), ..., selecting 3 and 7 and summing them (10), selecting 8 and 12 and summing them (20), ..., selecting 11 and 15 and summing them (26)).

I had thought doing that with the apparently natural:

`for i in arra[i, j, k]:`

for j in arra[i, j, k]:

for k in arra[i, j, k]:

sum...

The problem is that Python certainly doesn't allow to do what I want in this way. If it were a 2D array it would be easier as I know that the iterator first runs through the rows, so you can transpose to run along the columns, but for a multidimensional (3D in this case) array (N, M, P) with N, M, P >> 1, I was wondering how it could be done.

Answer

You can use `map`

to get this done:

```
import numpy as np
arra = np.arange(16).reshape(2, 2, 4)
```

Then the command:

```
map(sum, arra)
```

gives you the desired output:

```
[array([ 4, 6, 8, 10]), array([20, 22, 24, 26])]
```

Alternatively, you can also use a list comprehension:

```
res = [sum(ai) for ai in arra]
```

Then `res`

looks like this:

```
[array([ 4, 6, 8, 10]), array([20, 22, 24, 26])]
```

**EDIT:**

If you want to add identical rows - as you mentioned in the comments below this answer - you can do (using `zip`

):

```
map(sum, zip(*arra))
```

which gives you the desired output:

```
[array([ 8, 10, 12, 14]), array([16, 18, 20, 22])]
```

For the sake of completeness also the list comprehension:

```
[sum(ai) for ai in zip(*arra)]
```

which gives you the same output.

Source (Stackoverflow)

Comments