I've beeen taking deep learning course on coursera. While I was doing my assignment I saw one piece of code on the github.
1. numpyArr[...,c]
2. numpyArr[:,:,:,c]
What is the difference between these slicing methods?
Asked
Active
Viewed 73 times
0
smci
- 32,567
- 20
- 113
- 146
Omer Beden
- 31
- 5
-
1Not valid syntax – make sure that `mylist` really **is** a `list`. – meowgoesthedog Apr 24 '19 at 08:46
-
Could you link to the specific Github page where you saw the piece of code? – Plenka Apr 24 '19 at 08:49
-
The second is about multidimensional arrays, right? `mylist [:,:,:,c]` is actually a 4D array, not a list. – smci Apr 24 '19 at 08:51
-
https://github.com/Kulbear/deep-learning-coursera/blob/master/Convolutional%20Neural%20Networks/Convolution%20model%20-%20Step%20by%20Step%20-%20v1.ipynb – Omer Beden Apr 24 '19 at 08:51
-
2@meowgoesthedog The *syntax* is perfectly fine. The *meaning* doesn't work for `list` though. – deceze Apr 24 '19 at 08:51
-
yes, multidimensional arrays – Omer Beden Apr 24 '19 at 08:51
-
It's specific to [tag:numpy] then. – deceze Apr 24 '19 at 08:52
-
@ömerbeden: actually **numpy's ndarrays**, not Python's native multidimensional arrays. – smci Apr 24 '19 at 08:53
-
ok, I got that's specific for numpy then, Is there any difference ? – Omer Beden Apr 24 '19 at 08:57
-
The second syntax `[:,:,:,c]` is the more commonly used, I am seeing the other syntax for the first time – anand_v.singh Apr 24 '19 at 09:01
-
@anand_v.singh Thanks a lot. Actually I 'm seeing the other syntax first ,too. In the practice I can't find any difference between those methods. – Omer Beden Apr 24 '19 at 09:03
-
Also using the first syntax `[…,c]` I was unable to figure out a way to select a certain value from an intermediate dimension, I am now sure if it exists, hence I am not going to add an answer. i.e. I was unable to find an equivalent of `[:,:,c,:]` – anand_v.singh Apr 24 '19 at 09:08
-
@anand_v.singh it is `arr[...,c,:]` – oekopez Apr 24 '19 at 09:15
-
@oekopez so you can get till one dimension by `,:` and use elipses to include all the remaining dimensions on either side but only one side, is my understanding correct? – anand_v.singh Apr 24 '19 at 09:17
1 Answers
4
If both arrays have 4 dimensions there is no difference in the result. However, if you do not really care about the number of dimensions, using the Ellipsis (...) just indicates any number of dimensions. So the first version means:
"get all dimensions but from the last one (whatever the last is) only entry c "
and the second means
"get dimensions 0, 1, 2 complete and from dimension 3 only entry c.
Which is the same for a 4-d array but different for a 5d array.
For an array with many dimensions even more fun is possible:
arr = np.random.uniform(size=(3, 3, 3, 3, 3))
print(arr[1, ..., 2, 3].shape)
Which means: get the second entry on the first dimension and of that entry 2 and 3 of the two last dimensions with everything in between.
Some years ago, this has already been asked, but one needs to know that ... is the Ellipsis.
oekopez
- 1,080
- 1
- 11
- 19