once more unto the decomposition breach, dear friends! #27212

Sacha0 · 2018-05-23T01:46:02Z

...or close up the wall with our obscure initialisms!

This pull request is take three on the recent decompositions rework; take two was #27159, and take one was #26997.

Specifically, this pull request deprecates lufact[!], schurfact[!], lqfact[!], qrfact[!], ldltfact[!], and svdfact[!] to lu[!], schur[!], lq[!], qr[!], ldlt[!], and svd[!] respectively, clobbering existing methods of the latter functions.

Additionally, this pull request (will eventually) deprecate eig/eigfact[!] to eigen[!], chol[!]/cholfact[!] to cholesky[!], bkfact[!] to bunchkaufman[!], and hessfact[!] to hessenberg[!].

To facilitate this deprecation, this pull request adds destructuring via iteration to some of the decomposition objects, and (immediately deprecated) getindex(S::($DecompositionType), i::Integer) methods that yield the decomposition components produced by iteration.

Pending items:

Deprecate bkfact[!] to bunchkaufman[!]
Deprecate hessfact[!] to hessenberg[!]
Deprecate eigfact[!] to eigen[!]
Deprecate cholfact[!] to cholesky[!]
Deprecate eig to eigen
Deprecate chol[!] to cholesky[!]
Deprecate transitional getindex methods

Best!

Sacha0 · 2018-05-23T05:15:13Z

To my surprise and delight, I was able to finish everything this evening except for the gentler chol[!] -> cholesky[!] deprecation. I hope to eke out time for any issues CI catches tomorrow morning, and will try for the preceding last deprecation then as well. If I don't manage that last commit prior to other obligations, I will try to eke out additional time between those obligations, but otherwise pass the torch to @mbauman till Thursday :).

fredrikekre · 2018-05-23T11:28:25Z

Feel free to cherry pick from https://github.com/JuliaLang/julia/commits/breakmore

mbauman · 2018-05-23T14:37:57Z

stdlib/LinearAlgebra/src/deprecated.jl

@@ -1,4 +1,4 @@
-# This file is a part of Julia. License is MIT: https://julialang.org/license
+ # This file is a part of Julia. License is MIT: https://julialang.org/license


Good catch! Should be fixed :).

mbauman · 2018-05-23T14:39:03Z

stdlib/LinearAlgebra/src/deprecated.jl

@@ -7,6 +7,7 @@ using Base: @deprecate, depwarn
 @deprecate cond(F::LinearAlgebra.LU, p::Integer) cond(convert(AbstractArray, F), p)

 # PR #22188
+export cholfact, cholfact!


The @deprecate macro does the exporting for you by default

Indeed so :). I had forgotten at first, and once I remembered halfway through I decided to proceed consistently. Please feel welcome to remove the explicit exports!

Not a major concern by any means — just thought I'd mention it as I was reviewing.

mbauman · 2018-05-23T14:48:03Z

stdlib/LinearAlgebra/src/deprecated.jl

+# deprecate eig(...) in favor of eigfact and factorization destructuring
+export eig
+function eig(A::Union{Number, StridedMatrix}; permute::Bool=true, scale::Bool=true)
+    depwarn(string("`eig(A[, permute, scale])` has been deprecated in favor of ",


I find this eig(A[, permute, scale]) notation a little confusing — maybe just literally `eig(A; ...)`?

mbauman · 2018-05-23T14:49:28Z

stdlib/LinearAlgebra/src/deprecated.jl

+end
+
+# deprecate transitional decomposition getindex methods out of the blocks
+@inline function Base.getindex(S::LU, i::Integer)


This (and below) cannot be inlined for the deprecation to work. depwarn needs to have a stack frame to do its warn-once tracking.

mbauman · 2018-05-23T14:56:25Z

stdlib/LinearAlgebra/src/hessenberg.jl

 Hessenberg(A::StridedMatrix) = Hessenberg(LAPACK.gehrd!(A)...)

+# iteration for destructuring into components
+Base.iterate(S::Hessenberg) = (S.Q, Val(:H))
+Base.iterate(S::Hessenberg, ::Val{:H}) = (S.H, Val(:done))


This (and the others) is quite clever. I'm still catching up with all the things the new optimizer can do, but this seems like it'll be out of its reach and add quite a bit of dynamic dispatch. Maybe just:

+Base.iterate(S::Hessenberg) = (S.Q, false) +Base.iterate(S::Hessenberg, isdone) = isdone ? nothing : (S.H, true)

Or for the bigger factorizations, this might have a better chance of piggy-backing on the inference magic for tuples (it doesn't quite get it right now, though):

@inline function Base.iterate(x::S, state=nothing) tup = (x.a, x.b, x.c) state===nothing ? iterate(tup) : iterate(tup, state) end

Sacha0 · 2018-05-23T16:41:32Z

(Much thanks @fredrikekre and @mbauman! :D)

Sacha0 · 2018-05-23T17:00:56Z

Incorporated Matt's getindex @inline removals into the relevant earlier commit, fixed the one CI failure in the SparseArrays suite, and removed the errant whitespace Matt found, so all commits up to "Expand pull request numbers for some news items" should be passing CI now.

Had a brief look at one of Fred's commits: I am not certain that sqrt is generally substitutable for chol in choltype, hence in the relevant commit of #27159 I kept chol[!] around internally as legacychol[!] for the moment. Might be overly conservative though :).

Out of time now, hope to find bits of time again in a couple of hours. Thanks all and best!

simonbyrne · 2018-05-23T20:47:25Z

It should be possible to deprecate chol(X) to cholesky(X).U

fredrikekre · 2018-05-23T21:18:46Z

I am not certain that sqrt is generally substitutable for chol in choltype

Maybe not, but we should only have to keep the legacy_chol method for Number, since choltype calls promote_type(..., Float32) which only makes sense for Numbers (right?). I think it boils down to redefining choltype as:

function choltype(A)
    x = one(eltype(A))
    promote_type(typeof(x), typeof(sqrt(abs(real(x)))), Float32)
end

and then we don't need chol or legacychol.

It should be possible to deprecate chol(X) to cholesky(X).U

62e0b7e

StefanKarpinski · 2018-05-24T01:37:27Z

Failures are related to Pkg3 being renamed and/or timeouts.

Jutho · 2018-08-21T08:26:15Z

I very much like the new factorization interface and all the work that has been done here. Congrats to everbody involved.

I do have one question. I tried to look for a discussion about the name change eig to eigen. Why was this decided or deemed necessary? svd is also a three letter word so that can be it. I like the name changes from chol to cholesky and the full name hessenberg, but eig seems so standard. It's also used by Matlab and Numpy. In fact, if it were not for eigen, the names would be identical to Numpy/Scipy: numpy.linalg.eig, numpy.linalg.svd, numpy.linalg.cholesky, scipy.linalg.hessenberg, scipy.linalg.schur.

nalimilan · 2018-09-03T08:10:11Z

See also https://discourse.julialang.org/t/eigen-the-odd-duck/14451.

KristofferC · 2019-09-30T12:19:05Z

There are still no docs for iterate on factorization objects?

once more unto the decomposition breach, dear friends!

Sacha0 added 12 commits May 22, 2018 17:59

Deprecate lufact to lu.

6773f2d

Deprecate schurfact to schur.

8cc18fb

Deprecate lqfact to lq.

a9306ef

Deprecate qrfact to qr.

ac57d3e

Deprecate ldltfact to ldlt.

528de42

Deprecate lufact! to lu!.

62b1337

Deprecate schurfact! to schur!.

c8b4ac9

Deprecate lqfact! to lq!.

099bb32

Deprecate qrfact! to qr!.

23c8ff5

Deprecate ldltfact! to ldlt!.

a659ee9

Deprecate svdfact! to svd!.

1084fb4

Deprecate svdfact to svd.

0eda8e4

Sacha0 added linear algebra Linear algebra deprecation This change introduces or involves a deprecation labels May 23, 2018

Sacha0 added this to the 1.0 milestone May 23, 2018

Sacha0 mentioned this pull request May 23, 2018

break all the decompositions #27159

Closed

Sacha0 added 5 commits May 22, 2018 19:25

Deprecate bkfact to bunchkaufman.

a729118

Deprecate bkfact! to bunchkaufman!.

cf08aec

Deprecate hessfact to hessenberg.

5797f88

Deprecate hessfact! to hessenberg!.

ab672a6

Deprecate eigfact to eigen.

0d3c2ea

Sacha0 force-pushed the breakless branch 2 times, most recently from 792170c to 6927882 Compare May 23, 2018 04:13

Sacha0 added 3 commits May 22, 2018 21:25

Deprecate eigfact! to eigen!.

ee57c76

Deprecate cholfact to cholesky.

8f2b195

Deprecate cholfact! to cholesky!.

6e5d700

Sacha0 force-pushed the breakless branch from 6927882 to 51b1e3b Compare May 23, 2018 04:45

JeffBezanson removed this from the 1.0 milestone May 23, 2018

JeffBezanson added this to the 0.7 milestone May 23, 2018

mbauman reviewed May 23, 2018

View reviewed changes

Deprecate eig to eigen and destructuring via iteration.

aa65388

Sacha0 force-pushed the breakless branch from ca670f0 to f268aa9 Compare May 23, 2018 16:53

Sacha0 added 2 commits May 23, 2018 09:56

Deprecate decomposition getindex methods in favor of property accessors.

4873d97

Expand pull request numbers for some news items.

edf5bd2

Sacha0 force-pushed the breakless branch from f268aa9 to edf5bd2 Compare May 23, 2018 16:56

StefanKarpinski merged commit 2f93c94 into JuliaLang:master May 24, 2018

This was referenced May 24, 2018

Don't test that Pkg exports add JuliaLang/Compat.jl#555

Merged

Fate of qr JuliaLang/Compat.jl#556

Closed

mbauman mentioned this pull request May 24, 2018

Finish breaking the decompositions: deprecate chol and chol! #27249

Merged

Sacha0 deleted the breakless branch May 24, 2018 19:01

Sacha0 mentioned this pull request May 24, 2018

revisit decomposition iteration mechanism #27250

Open

andreasnoack mentioned this pull request Jun 5, 2018

Justification for eig, eigvecs, eigvals and eigfact? #15573

Closed

garrison mentioned this pull request Aug 3, 2018

qr(A; full=false) no longer available? #27397

Closed

Keno pushed a commit that referenced this pull request Jun 5, 2024

Merge pull request #27212 from Sacha0/breakless

d2f19fd

once more unto the decomposition breach, dear friends!

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

once more unto the decomposition breach, dear friends! #27212

once more unto the decomposition breach, dear friends! #27212

Sacha0 commented May 23, 2018 •

edited by mbauman

Loading

Sacha0 commented May 23, 2018 •

edited

Loading

fredrikekre commented May 23, 2018

mbauman May 23, 2018

Sacha0 May 23, 2018

mbauman May 23, 2018

Sacha0 May 23, 2018

mbauman May 23, 2018

mbauman May 23, 2018

mbauman May 23, 2018

mbauman May 23, 2018

Sacha0 commented May 23, 2018

Sacha0 commented May 23, 2018

simonbyrne commented May 23, 2018

fredrikekre commented May 23, 2018

StefanKarpinski commented May 24, 2018

Jutho commented Aug 21, 2018

nalimilan commented Sep 3, 2018

KristofferC commented Sep 30, 2019

		@@ -1,4 +1,4 @@
		# This file is a part of Julia. License is MIT: https://julialang.org/license
		# This file is a part of Julia. License is MIT: https://julialang.org/license

once more unto the decomposition breach, dear friends! #27212

once more unto the decomposition breach, dear friends! #27212

Conversation

Sacha0 commented May 23, 2018 • edited by mbauman Loading

Sacha0 commented May 23, 2018 • edited Loading

fredrikekre commented May 23, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Sacha0 commented May 23, 2018

Sacha0 commented May 23, 2018

simonbyrne commented May 23, 2018

fredrikekre commented May 23, 2018

StefanKarpinski commented May 24, 2018

Jutho commented Aug 21, 2018

nalimilan commented Sep 3, 2018

KristofferC commented Sep 30, 2019

Sacha0 commented May 23, 2018 •

edited by mbauman

Loading

Sacha0 commented May 23, 2018 •

edited

Loading