MAINT: sparse: reformat str and repr for sparse arrays, correct 1D coords, improve dtype looks #20649
Conversation
github-actions
bot
added
scipy.sparse
maintenance
|
When changing the str/repr format, I didn't change the refguide doctests to match. The CI reports 47 doctests that need to be updated. I've updated/corrected them. I also changed the term "type" to "dtype" because that is what we are showing... not idx_type or the object's type. |
scipy/sparse/_lil.py
Outdated
| if self.nnz == 0: | ||
| return val | ||
|
|
||
| val += '\n Coords\tValue\n' |
There was a problem hiding this comment.
The base class uses Coords and Values, so we should use Values here too.
Though I wonder: since this doesn't support maxprint, should we even have this override?
There was a problem hiding this comment.
Good point -- This str override could print something very long. And the base class will handle it just fine.
I've removed that method from _lilbase here.
I've also rebased to include the csr-1d stuff in case that changes anything.
|
The CI failure seems not to be related to this PR. I believe this |
|
Let me restart the failing CI job for |
|
Good thanks! Rerunning the test seems to have let it pass. :) |
| return '\n'.join([(' {}\t{}'.format(*t)) for t in triples]) | ||
| def tostr(coords, data): | ||
| pairs = zip(zip(*coords), data) | ||
| return '\n'.join(f' {idx}\t{val}' for idx, val in pairs) |
There was a problem hiding this comment.
Note: in Numpy 2.0+, this will format the indices with np.int32() around each one. For example:
(np.int32(3), np.int32(1)) 0.2435
|
The "add unittests" commit adds tests that expose the change in str for numpy 2.0. (somehow doctests are not being tested using numpy 2.0 yet) The "fix numpy 2 printing" commit converts values to python numeric before printing (using If these tests pass I think we're good to start a re-review by @perimosocordiae due to the numpy changes. |
Fixes #20378
Fixes #20377
Currently,
str(A)for sparse A with no stored elements returns the empty string.In
repr(A), for 1D arrays, the shape does not appear nicely, e.g. "958 sparse array of type ...". 2D is "4x3 sparse array of type".This PR puts in repr:
formatanddtypein first line,nnzandshapein 2nd line.For
str, the text from repr is the start, followed by coord/value column titles followed by coord/value info as before, but now it works for 1D. Also, if no stored values exist, thereprtext is returned.