ENH: Added fill option to np.atleast_2d and np.atleast_3d
#25170
Conversation
|
Can someone help me fix the failed build / give some feedback? |
fill option to np.atleast_2d and np.atleast_3dfill option to np.atleast_2d and np.atleast_3d
|
Up |
fd4913c to
7e09862
Compare
|
Dear, @melissawm @charris, thank you both for having shown interest at some point. Do you think it is possible to get this PR reviewed or even discussed at some point? What would the procedure be? Because I do feel a bit unconfortable directly pinging you for attention. But I don't know any solution to get it out of the abyss! Thanks! |
|
Somewhat related? data-apis/array-api#494 |
|
We discussed this in a recent triage meeting. Since there is no corresponding issue, could you explain the motivation? I did see #25805 about consistency with left-right and |
Thank you for your comment. I do think that new_ary = np.atleast_nd(ary, nd, fill)with fill (str | Iterable[int]): Where to add missing dimensions, if any. If
a string, it must be one of ``{'left', 'right'}``. Otherwise, it
must be a permutation of ``np.arange(nd)``, in which case missing
dimensions will be added in axes indexed by ``fill[:nd-ary.ndim]``.This could be even more permissive in a few ways. For example Also, still with The current implementations of np.atleast_nd(ary, 2, fill='left') # current `atleast_2d`
np.atleast_nd(ary, 3, fill=(2, 0, 1)) # current `atleast_3d`
np.atleast_nd(ary, 3, fill=(-1, 'left')) # another possible syntax for current `atleast_3d`For my PR, I only focused on a small modification, to avoid going through a long discussion / approval process required to add a new method. The use cases can be plenty. A simple one would be to allow a user to call a function treating 2d arrays by passing either one 2d array or a batch of 2d arrays (so a 3d tensor with This can feel silly, but I'm pretty sure there numerous use cases, especially with Thanks! |
|
Any feedback? I think it is reasonable to merge this as it is a very very light and compatible change, providing more flexibility. As seen above, the implementation of |
|
Closing in favor of finishing #18386 and making it compliant with the array API. Please reopen if you wish to pursue this as well. |
Closes #25805 since it provides override option.
Added optional keyword argument
filltonp.atleast_2dandnp.atleast_3d, to specify where the missing dimension(s) should be added.np.atleast_1dfor obvious reasons, but if signature consistency is an issue, it can be done,This can be useful when working with Fortan convention for example, or for locally-specifi reasons.
Illustrative examples: