fix doc for the op sort() and argsort() #15317
Conversation
|
@mxnet-label-bot add [Doc, pr-awaiting-review] |
| @@ -114,7 +114,7 @@ Examples:: | |||
| [ 1., 3.]] | |||
|
|
|||
| // flattens and then sorts | |||
| sort(x) = [ 1., 1., 3., 4.] | |||
| sort(x, axis=None) = [ 1., 1., 3., 4.] | |||
There was a problem hiding this comment.
This sort makes sense.
But I'm thrown off by the other example. Sorting along the last axis would be 3 then 4 unless it is descending rather than ascending. Does this need to be specified for clarity?
There was a problem hiding this comment.
Hi,
Thanks for your reply!
Do you mean this example?
// sorts along the last axis
sort(x) = [[ 1., 4.], [ 1., 3.]]
Since x = [[ 1, 4],[ 3, 1]], the pairs in the last axis are [1,4] and [3,1]. Sorting only along the last axis in a descend way would be [1,4] and [1,3]. So, in my opinion, it is correct? or maybe I'm wrong?
There was a problem hiding this comment.
I tested it in numpy and it matches fine.
There was a problem hiding this comment.
Sorry that I didn't get your point. What I did wrong in this PR?
If the argument axis=None is not given, this function will sort the array alone the last axis. However, in the second example, we want the array to be flattened and then sorted.
There was a problem hiding this comment.
Everything is fine. I might have needed more examples to see the patterns, but I understand that if you're conversant with NumPy then this all works as expected.
|
@mxnet-label-bot add [pr-awaiting-merge] is this good to go? |
Description
Errors in the example of
sort()andargsort(). Those op flatten and sort the array only whenaxis=Noneis given explicitly.Checklist