New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
MAINT: Backports for 1.4.0rc2 #11161
MAINT: Backports for 1.4.0rc2 #11161
Conversation
* set version to 1.4.0rc2 unreleased
It is causing some errors in documentation build due to unused references
change a few size/index related variables from int -> int_p disable wraparound and boundscheck in a few other places
This is the non-default code path, not realised usually.
The issue could occur for CSR matrices with row start index > stop index. This fixes the same issue for column slice of CSC matrices.
* add out-of-bounds and invalid sparse indexing test cases requested by reviewer * adjust compressed.py source code such that start index > stop index case now returns an empty object (of same class as input)
This reverts commit 1e870bb.
* account for backported PRs: 11081, 11096, 11124, 11126, 11127
* BUG: Fix random state bug multiscale_graphcorr * BUG: Fix index error in multiscale_graphcorr * BUG: fix nonetype exception * ENH: Remove workers dependence for random_state in multiscale_graphcorr * MAIN: Move random state keyword to end of parameter list * MAIN: change list comprehension so it is more clear * MAIN: fix typo in seed parameter * ENH: Change seed generation in multiscale_graphcorr * DOC: Use random_state keyword in example * MAIN: Add whitespace around bitwise operator
* BUG: empty sparse slice shapes * CSC and CSR sparse matrix slices that return empty matrices should now correctly preserve a shape of (0, original_shape[1]) * refactor _process_slice replace use of 'empty' with i0=i1 which gives an empty slice this will keep the existing logic for shape handling * TST: add additional CSR and CSC slicing test cases
Updated with more backports on Dec. 3/2019 & corresponding release notes adjustments. The list of backports in the original description has been updated for a quick glance. If CI is all green after this I'll be pretty tempted to merge & move forward with |
@@ -392,6 +391,7 @@ Authors | |||
======= | |||
|
|||
* @endolith | |||
* @wenhui-prudencemed + |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@wenhui-prudencemed I've added your GitHub handle to the updated release notes---apologies in advance if you have another preference for naming & we end up merging first (fairly likely considering recent delays!).
Azure failure is Python 3.8/ Windows:
Maybe that's a fluke (@lobpcg ?). Travis failure is related to matrix entries pulling NumPy pre-release with the change I never really liked:
@rgommers @mattip @charris Should I just pin the NumPy versions for the SciPy |
The ragged array related deprecation warnings seem to be a mix between things that should be cleaned up in our code (e.g. the We'll discuss it in an hour in the triage call, for now just ignore the failures I'd say @tylerjereddy. No point messing with CI pinning, either we'll clean it all up asap or we'll revert the change in NumPy for a while. |
The These are the relevant lines:
Full traceback:
EDIT: that's a crash in |
It is possible for LOBPCG to crash, in principle, although that is very rare. I do not recollect seeing it crashing in this test. Please try to re-run just with a different seed in https://github.com/scipy/scipy/blob/master/scipy/sparse/linalg/eigen/lobpcg/tests/test_lobpcg.py line 209 |
I ran the test 1000 times with the same/original seed using
It was always successful: I guess I'll try with Python 3.8 now. Doing it on Windows may take a bit more work for me locally. |
Stable after 2500 repeats on Python 3.8/Linux. For Windows maybe I will check on my Azure fork tomorrow by replacing the normal test suite with just a large repeat count of the test in question, if we want to be safe. |
I did those tests on latest master though.. |
Unless the hardware is faulty, the repeats must be absolutely identical, with all the digits, i.e. no reason for repeats. |
Alright, close/reopen to flush CI again. Hopefully the If I'm not mistaken we may need to be patient for the |
I'm not sure if this is the right PR in which to report this, but I just noticed that the addition of Also in the release notes, a blank line is needed before
to separate it from the previous item. |
|
* update 1.4.0 release notes to reflect backported PRs: 11152, 11166, 11167 * remove a redundant mention of multiscale_graphcorr in the release notes and improve spacing around the original mention of multiscale_graphcorr
b8c1715
to
7bd5ad6
Compare
I addressed Warren's release notes corrections and pushed those in---this should also serve to flush CI and tell us if nightly NumPy wheels will give us a green light now. |
Ok, still have the dtype=object problem, for now. |
I just triggered a wheels build of master, the last was two days ago. |
Reran the failing tests, they are passing now. |
Great, thanks Chuck, merging backport PR with CI all green. |
Contains:
1.4.0rc2
unreleasedCurrently Blocked By:
BUG: Fix random state bug multiscale_graphcorr #11152BUG: empty sparse slice shapes #11166