META: deprecation tracker

[h-vetinari]

Please edit this tracker as necessary

Below a list of deprecations by kind and scipy release. Ideally all deprecations should come with an announced behaviour change, and with an announced timeframe (i.e. SciPy version where things will be changed/removed), and other items migrated there as appropriate.

In most cases, we aim for removal/behaviour change one year after a given deprecation is introduced in 1.N (not just in documentation, but in a way that the user cannot miss, i.e. raise DeprecationWarning upon use). Due to the 6 month release cadence this often ends up being 1.(N+2), but exceptional out-of-band releases can occur, so try to avoid promising exact versions. For more disruptive cases, adaptations (e.g. longer timeframes) can be made as necessary.

Note
If you add newly-deprecated functionality to the tracker, please do so twice - once for tracking what we did for the upcoming release (whether following up on old doc-only deprecation or completely new deprecations), and once for tracking when that deprecation is scheduled to be executed (generally two releases later).

Cleaning up the past

Deprecations in documentation only

None currently tracked, if you find one, let us know!

Deprecated without announced removal

None currently tracked, if you find one, let us know!

Removal announced, missed

• 1.5 (announced removal for 1.7)
  • ENH: special: update nbdtr and friends to use Boost implementation #15596
• 1.9 (announced removal for 1.11)
  • DEP: remove deprecated linprog-methods #15707

In progress for upcoming release

Note
For removals and new or sharpened deprecations, please keep the release notes up-to-date.

Old deprecations, warning sharpened for removal in 1.14/1.15

None so far

New deprecations

• DEP: deprecate positional arguments for signatures being changed in 1.14 cycle #18703
• DEP: deprecate and remove remaining usages of slur-adjacent "trapz" #20486
• DEP: special.comb: deprecate exact=True for non-integers #15722

Removal announced for 1.14

Open issues for these once 1.13 branches off of main.

• 0.11 (version where original deprecation was added)
  • DEP: switch sparse methods to kwarg-only; remove tol/restrt kwargs #20498 (follow-up to this & this)
• 0.19 (as above...)
  • DEP: stats: switch kendalltau to kwarg-only, remove initial_lexsort kwarg #20550 (follow-up to this & this)
• 1.0
  • DEP: signal: remove nyq / Hz kwargs in firwin* and switch to kwarg-only #20578 (follow-up to this & this)
• 1.5
  • DEP: linalg: remove turbo / eigvals kwargs from linalg.{eigh,eigvalsh} and switch to kwarg-only #20560 (follow up to this and this)
• 1.6
  • DEP: integrate: remove simps,trapz,cumtrapz #20278 (follow-up to this)
• 1.7
  • DEP: linalg: remove cond / rcond kwargs from linalg.pinv and switch to kwarg-only #20565 (follow-up of this and this)
• 1.10
  • DEP: replace interp2d by stub #20497 (for DEP: replace scipy.interpolate.interp2d by a stub #18583)
• 1.11
  • DEP: special: remove legacy kwarg from special.comb and switch to kwarg-only #20588 (basically this + this; delayed from 1.13 to 1.14)
  • DEP: integrate: switch simpson to kwarg-only, remove even kwarg #20554 (follow-up of this & this; delayed from 1.13 to 1.14)
  • DEP: signal: raise error using medfilt and order_filter with float128 and object dtypes #19673 (follow-up of this)
• 1.12
  • MAINT: remove incidental imports from private modules #19904
  • DEP: execute sparse array API deprecations #20499 (follow-up of this)
  • DEP: switch sparse methods to kwarg-only; remove tol/restrt kwargs #20498 (follow up to this, this and this)

Scheduled removals in future versions

Removal announced for 1.15

Open issues for these once 1.14 branches off of main.

• 1.12 (removal intentionally delayed by one release due to out-of-band 1.13)
  • raise ValueError for initial not in (0, None) in cumulative_trapezoid (follow-up to this)
  • remove scipy.signal.cmplx_sort (follow-up to this)
  • remove scipy.stats.rvs_ratio_uniforms (follow up to this)
  • remove wavelet functions (follow-up to this)
  • remove quadrature and romberg functions (follow-up to this)

• 1.13
  • raise error for factorial(<non-integer-scalar>, exact=True) (follow-up to this)
  • raise error for complex dtypes for PchipInterpolator and Akima1DInterpolator (follow up of DEP: deprecate complex dtypes in PchipInterpolator and Akima1DInterpolator #19752)

Removal announced for 1.16

Open issues for these once 1.15 branches off of main.

• 1.14
  • remove conjtransp() from dok_array/matrix classes (follow up of DEP: sparse: deprecate conjtransp() method for dok_array/matrix classes #20283)

Removal announced/planned for 2.0

• all private modules affected by Making clearer what namespaces are public by use of underscores #14360
• remove scipy.misc (follow-up of this)

Other

Outstanding deprecations

List most likely incomplete

• Deprecation of scipy.spatial.kdtree distance functions #10419
• Deprecate sparse.spmatrix.todense #14494
• DEP: stats: deprecate support for _rvs using _random_state. #15924

Outstanding behaviour changes

List incomplete (and intentionally so; not every tiny planned change needs to be here from the get go; but do add it if desired, ideally already above with a direct action item).

• 'sokalmichener' and 'rogerstanimoto' distances are the same (Trac #1486) #2011
• Filter design functions should accept other input than TF. #3259
• Sparse: document and test handling of explicit zeros #3343
• Support binary operators on binary sparse matrices #3590
• scipy.sparse Unexpected behaviour for numpy.dot(sparse,dense) and no support for output array #4063 / wishlist: sparse in-place AXPY/GEMM instead of deprecated sparsetools.csc_matvecs #5348
• numpy and scipy qr have different mode names. #4818
• scipy.signal needs a unified filter API #6137
• remove KDTree Python implementation #8923
• API: Inconsistent naming of arguments in svd / svds / eig / eigs #11334
• Hamming distance returns unexpected value for string comparisons #11991
• ENH: add more aka for cdist/pdist #14039
• ENH: refactor RNG usage to use np.random.Generator #14322
• ENH/DEP: Clean up quad return types #18574
• Tolerance arguments for scipy.sparse.linalg.{lsmr,lsqr} do not match related functions #19704

Reference

Closed but relevant for continued maintenance

• API: keywords only arguments #14714

[tupui]

Hi @h-vetinari, thanks for compiling this list! I would not have created issues for all of them, but kept a list and use GH's features to then create/link PRs when something is on the way. This to prevent spamming people following the repo.

[h-vetinari]

Understood, though I think text in an issue somewhere quickly gets forgotten (and discussing individual cases in a meta/mega-tracker quickly gets impossible to follow).

OTOH, if it's an issue in a milestone, many more people will look at it, with a much higher chance that it gets done (also, we can set "good first issue" labels!). I'm sorry about the spam; at least it should be (have been) a one-time thing; even if we continued to raise individual issues, it'd be incremental from now on.

[rgommers]

This is very useful, and will help going through and cleaning this up, thanks @h-vetinari. One minor comment: the abbreviation we use is DEP: rather than DEPR:. For consistency, it may be useful to remove that now, otherwise we'll end up with mixed acronyms long-term.

[h-vetinari]

One minor comment: the abbreviation we use is DEP: rather than DEPR:

Sorry about that. My first instinct when choosing such acronyms is to avoid ambiguity (such as "dep" for depencency management), but I didn't know that this was standardized already (is there a list somewhere?). In any case, I can change this in the issue titles (of those I raised). If we're consistent with the deprecate label, that schism will be even less of a problem.

[tupui]

... (is there a list somewhere?).

Yes the list is here https://scipy.github.io/devdocs/dev/contributor/development_workflow.html#writing-the-commit-message

[h-vetinari]

Great thanks! (I was looking through the templates and the wiki here...)

[j-bowhay]

@h-vetinari another warning that doesn't state version to be removed in https://github.com/scipy/scipy/blame/main/scipy/io/harwell_boeing.py

[h-vetinari]

Thanks @j-bowhay. I've excluded all the module-level deprecations that came from the work in #14360 above; I think we should track a timeline for removal, but I'm not sure when that's planned...

[h-vetinari]

I did another search for "deprecated" and found some more things that somehow did not show up in the github code search the last time. I added respective issues to the tracker.

[h-vetinari]

Now that 1.9 has branched, I removed the work done for that release from the tracker. Here it is for the purpose of record-keeping:

Deprecations that were executed for 1.9

• 0.13
  • DEP: remove object arrays in sparse matrices #15764
• 0.17
  • DEP: remove LAPACK *gegv functions #15758 (see also Deprecated routines in Netlib LAPACK >3.5.0 #5266)
• 0.18
  • DEP: remove ftol/xtol from neldermead #15751
• 0.19
  • DEP: deal with deprecation of ndim >1 in bspline #15734
  • DEP: Remove debug keyword from scipy.linalg.solve_* #15736
  • DEP: remove right keyword from interpolate.PPoly.extend #15752
  • DEP: Remove mlab mode from dendrogram #15754
  • DEP: remove old BSR methods #15759
• 1.0
  • DEP: remove spatial.distance.matching #15741
  • DEP: raise if fillvalue cannot be cast to output type in signal.convolve2d #15742
  • DEP: remove _ppform #15753
  • DEP: remove py_vq2  #15760
• 1.1
  • DEP: execute deprecation of inexact indices into sparse matrices #15732
  • DEP: remove signal.windows.hanning #15745
  • DEP: remove stats.itemfreq #15761
• 1.3
  • DEP: docstring-related deprecations #15757
• 1.4
  • BUG: NumPy's random module should not be in the scipy namespace. #14889
  • DEP: enforce radius for spatial.SphericalVoronoi #15743
• 1.5
  • DEP: stats: remove support for _rvs without size parameter #15747
  • DEP: remove stats.median_absolute_deviation #15762
• 1.6
  • DEP: remove k=None from KDTree.query #15746 (removal had been announced for 1.8)
  • DEP: remove n_jobs from kdtree #15750 (removal had been announced for 1.8)
• 1.7
  • DEP: Execute deprecation of pinv2 #15737
  • DEP: Execute deprecation for squeezing input vectors in spatial.distance  #15740 (removal had been announced for 1.9)

Old deprecations, warning sharpened for removal in 1.10/1.11

• 0.10
  • DEP: extradoc in _distn_infrastructure #15748
• 0.12
  • DEP: add deprecation warning for vertices kwarg in qhull #16255
• 0.19
  • DEP: add actual DeprecationWarning for sym_pos-keyword of scipy.linalg.solve #15735
• 1.5
  • DEP: add deprecation warning to maxiter kwarg in _minimize_tnc #16254
• 1.8
  • DEP: sharpen deprecation of dual_annealing argument 'local_search_options' #15744
• 1.9
  • DEP: sharpen deprecation for >1-dim inputs in optimize.minimize #15739

[tupui]

Thank you again @h-vetinari, this is great work and very useful 👏

[ev-br]

Edited in the interp2d deprecation plan from gh-17180 and the related mailing list discussion. Please let me know and/or adjust the landing page if what I added is not right.

[j-bowhay]

Maybe this would be a nice issue to pin for better visibility?

[h-vetinari]

Since 1.10 just branched, here's the things we did for 1.10 (now removed from the tracker)

Removed in 1.10

• 0.12
  • DEP: Execute deprecation of scipy.linalg.blas.{clapack, flapack} #16251
• 0.15
  • DEP: remove quiet parameter from fitpack #15733 (deprecation itself was undone; hence no release note)
• 1.7
  • DEP: execute deprecation of cond/rcond kwarg in linalg.pinv #16395
• 1.8
  • DEP: remove tol from NumericalInverseHermite #15749
  • DEP: remove dual_annealing argument 'local_search_options' #15807

Deprecated in 1.10, removal announced for 2.0

• DEP: Remove scipy.misc submodule  #15608

Old deprecations, warning sharpened for removal in 1.11/1.12

• 0.11
  • DEP: add warning to sparse.gmres deprecated kwarg restrt #16392
• 0.18
  • DEP: deprecate stats.kendalltau kwarg initial_lexsort #16388
• 1.0
  • DEP: add deprecation warning for kwargs nyq / Hz in firwin* #16253
• 1.5
  • scale='raw' in iqr (see DEP: sharpen stats deprecations #16389)
  • DEP: add deprecation warnings to kwargs turbo / eigvals of linalg.eigh #16252
• 1.7
  • DEP: add deprecation warning for binom_test  #16256
• 1.9
  • n in stats.moment and alpha in stats.interval (see DEP: sharpen stats deprecations #16389)

[h-vetinari]

Since 1.11 just branched, here's the things we did for that release (now removed from the tracker)

Removed in 1.11

• 0.10
  • DEP: remove extradoc keyword in _distn_infrastructure #15818
• 0.12
  • DEP: remove vertices kwarg in qhull #16270
• 0.16
  • Non-uniform timestamps in lsim (see here & here)
• 0.19
  • DEP: remove sym_pos-keyword of scipy.linalg.solve #15829
• 1.5
  • DEP: remove maxiter kwarg in _minimize_tnc #16269
• 1.9
  • Switched default of legacy= kwarg in special.comb (see here)
  • DEP: remove deprecated parameters from stats distributions #15706
  • DEP: remove non-numeric array support in stats.mode #15756
  • DEP: raise on >1-dim inputs for optimize.minimize  #15808
  • DEP: remove deprecated gilbrat distribution #15988
  • DEP: remove MaximumFlowResult.residual #16014
  • DEP: remove scipy.spatial.distance.kulsinski #16271
  • removal of kwargs n in stats.moment and alpha in stats.interval (follow-up to this and this)
  • DEP: stats.mode: raise with non-numeric input #18290

Old deprecations, warning sharpened for removal in 1.12/1.13

• 0.16
  • Deprecate or discourage signal.step2 signal.impulse2 signal.lsim2 #5676
  • DEP: absorb lsim2 into lsim #15755
• 1.1
  • DEP: Warn on deprecated windows-import in base scipy.signal namespace #18484
• 1.3
  • DEP: deprecate multiple-ellipsis handling in sparse matrix indexing #18485

New deprecations

• 1.11
  • Deprecate scipy.signal.{bspline, quadratic, cubic} #17455
  • DEP: linalg: deprecate tri{,u,l} #18021
  • even keyword in scipy.integrate.simpson (see here)
  • DEP: signal: deprecate using medfilt and order_filter with float128 and object dtypes #18341
  • Clean up sparse array API #18507

[j-bowhay]

Work completed for this release:

Old deprecations, warning sharpened for removal in 1.13/1.14

• 0.11 (version where original deprecation was added)
  • DEP: deprecate positional arguments in sparse.gmres #18577
• 0.19 (as above...)
  • DEP: deprecate positional arguments in stats.kendalltau #18578
• 1.0
  • DEP: deprecate positional arguments in firwin* #18579
• 1.1
  • DEP: change default of atol in scipy.sparse.linalg.* #15738
• 1.6
  • DEP: deprecate aliased integration methods that are (close to) slurs #18702
• 1.8
  • BUG: deprecation warnings for private API are misleading #18279 (more a clarification, but noted for completeness)

New deprecations

• 1.12
  • Deprecate positional arguments for signatures being changed in 1.12 cycle, originally tracked here
    • DEP: deprecate positional arguments for some methods in signal._fir_f… #18932
    • DEP: deprecate positional arguments for special.comb #18933
    • DEP: deprecate positional arguments for some methods in sparse.linalg… #18934
    • DEP: deprecate positional arguments for linalg.pinv #18935
    • DEP: deprecate positional arguments for stats.kendalltau #18936
    • DEP: sparse.linalg: deprecate positional arguments for gcrotmk, lgmres, minres, tfqmr #18943
    • DEP: linalg: deprecate positional args for eigh/eigvalsh #18946
  • DEP: signal: deprecate cmplx_sort #18587
  • DEP: integrate: Deprecate initial values other than None or 0 in cumulative_trapezoid #18219
  • deprecate scipy.stats.rvs_ratio_uniforms (in MAINT: Move rvs_ratio_uniforms to sampling #18622)
  • DEP: deprecate scipy.signal wavelets #19340
  • DEP: integrate: deprecate romberg and quadrature #19510

Removed for 1.12

• 1.5 (version where original deprecation was added; raises DeprecationWarning as of 1.10)
  • DEP: removal of scale='raw' in iqr #18580
• 1.7
  • DEP: removal of stats.binom_test #18581
• 1.10
  • DEP: removal of parameter centered of stats.qmc.LatinHypercube #18582

[lucascolley]

@j-bowhay probably useful to have gh-18703 linked somewhere in the top comment?

[h-vetinari]

Work completed for this release:

Since you took care of most of the tracking this cycle, could you open a PR that ensures the release notes (already on the 1.12 branch) are up-to-date? For example, I see that not all expired deprecations are listed.

[lucascolley]

For executing the deprecations of things that were never supposed to be public like gh-19801 for 1.13.0, is that going to be done per module @j-bowhay ? I have no idea how many there are.

[j-bowhay]

I was planning to do it all at once, I think it should be a case of just tweaking the helper and removing things until the tests are happy.

[h-vetinari]

In #19880, there was some discussion around deprecations, as the "out-of-band" release of 1.13 throws a bit of a wrench into the deprecations as they've been laid out (on the basis of a 6 month release cadence).

@rgommers noted:

Please consider following https://numpy.org/neps/nep-0023-backwards-compatibility.html#implementing-deprecations-and-removals for future deprecations though, it works pretty well in NumPy.

We do follow that pretty much to a tee, with the exception that we started putting version numbers for scheduled removal in the warnings (that aspect does not seem mandated either way by NEP23), and that we don't have "is deprecated since" bit.

I do think the removal date is user-relevant (despite the fact that we don't know the future); by seeing the warning, they're obviously already on an affected version, so when it was introduced is less relevant.

That said, it's IMO fine to switch to a "deprecated as of" formulation, as that avoids the need for a crystal ball. That way we can always wait for a year before removal - though in turn, we shouldn't revert back to the bad old days of letting these deprecation warnings linger indefinitely.

---

For the here and now (and especially given that 1.12 hasn't been released yet), we should double-check current deprecations that we don't remove things too early/eagerly. Looking at the list above, a lot of them had been deprecated previously already (or we even extended the deprecation period due to also deprecating positional use), so the only ones where I think it makes sense to push the announced deprecation date from 1.14 to 1.15 (or reformulate the warning completely) are the following:

• new deprecations in 1.12:
  • DEP: signal: deprecate cmplx_sort #18587
  • DEP: integrate: Deprecate initial values other than None or 0 in cumulative_trapezoid #18219
  • deprecate scipy.stats.rvs_ratio_uniforms (in MAINT: Move rvs_ratio_uniforms to sampling #18622)
  • DEP: deprecate scipy.signal wavelets #19340
  • DEP: integrate: deprecate romberg and quadrature #19510

For things deprecated in 1.11 and already removed on main, the emerging consensus was to leave these things as-is. The only exceptions are #18583 (delayed to 1.14, see discussion in #18618), #19673, resp. the follow-up to #18507; for the latter two we should decide what we want to do; I think it'd be easiest just to wait.

I'll try to prepare a PR for the 1.12 backport.

[j-bowhay]

Work completed for the 1.13 release:

New deprecations

• 18bb04b
• DEP: deprecate complex dtypes in PchipInterpolator and Akima1DInterpolator #19752

Removal announced for 1.13

• DEP: signal: remove scipy.signal.{lsim2,impulse2,step2} #19675 (follow-up of this, this & this)
• DEP: signal: remove ability to import window functions from signal namespace #19676 (follow-up of this)
• MAINT: sparse: Stop supporting multi-Ellipsis indexing #18541 (follow-up of this)
• DEP: signal: remove scipy.signal.{bspline,quadratic,cubic} #19672 (follow-up of this)
• DEP: linalg: remove tri{,u,l} #19674 (follow-up of this)
• f5265e5 (follow-up of this)

[tylerjereddy]

@h-vetinari I proposed the SciPy 1.14.0 release schedule here: https://discuss.scientific-python.org/t/proposed-release-schedule-for-scipy-1-14-0/1151?u=tylerjereddy