Add a keepdims argument to linspace

@mhvk

Actually, I'm now wondering if we should add a keepdims=True to linspace, with semantics "replace the axisth axis with the new linspace and squeeze". @mhvk, do you have thoughts on this?

@eric-wieser - you mean something like the opposite of a reduction, where keepdims=True tells linspace that there is already a dimension of 1 inside start and stop? On the one hand, it doesn't quite feel as natural, since in that case one has to check that that dimension is indeed 1. On the other hand, I can see the nice aspect that then start and stop would broadcast with the result. Overall, the "other hand" wins for me...

Originally posted by @mhvk and @eric-wieser in #14209

Some justification for this - it would these to mean the same thing:

np.linspace(
    arr.max(axis=ax),
    arr.min(axis=ax),
    axis=ax
)

np.linspace(
    arr.max(axis=ax, keepdims=True),
    arr.min(axis=ax, keepdims=True),
    axis=ax, keepdims=True
)

which is handy for algorithms that already compute the keepdims version of max and min.

@eric-wieser

Hey @eric-wieser, I'm interested to take a look at this.

Just to make sure I understand the requirements.

arr = np.arange(4).reshape((2,2))
ax = 0
arr_min = arr.min(axis=ax, keepdims=True)  # ndim = 2
arr_max = arr.max(axis=ax, keepdims=True)  # ndim = 2
np.linspace(arr_min, arr_max, axis=ax)  # ndim = 3 != 2

So, we basically want to add a keep_dim parameter to np.linspace so that its output has the same dimension as arr_min and arr_max?

@efiring

I agree with @efiring that the usage of this new keyword might not very common and so does not worth the complexity it adds.

In addition, as pointed out by @shoyer, one can achieve the same goal with the following code:

np.linspace(
    arr.min(axis=ax, keepdims=True).squeeze(ax), 
    arr.max(axis=ax, keepdims=True).squeeze(ax), 
    axis=ax,
)

So, I'm closing this PR and the associated issue.

Please refer to the PR for more comments.

Thanks for the effort ZJ! Sometimes its necessary to propose a solution to get a resolution unfortunately.

No worries. It's a good learning opportunity for me. 😄

Sorry for the wild goose chase - I think the arguments made on the mailing list are pretty reasonable :)

Navigation Menu

Repository navigation

Add a keepdims argument to linspace #14425

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Add a keepdims argument to linspace #14425

Description

Activity

zjpoh commented on Nov 15, 2019

zjpoh commented on Dec 11, 2019

seberg commented on Dec 11, 2019

zjpoh commented on Dec 11, 2019

eric-wieser commented on Dec 11, 2019

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions