rht
commented
3 years ago You basically just need to use this: https://github.com/scipy/scipy/blob/master/scipy/optimize/lbfgsb_src/lbfgsb.f and recompile.
Open rht opened 3 years ago
rht
commented
3 years ago You basically just need to use this: https://github.com/scipy/scipy/blob/master/scipy/optimize/lbfgsb_src/lbfgsb.f and recompile.
Gnimuc
commented
3 years ago There is https://github.com/JuliaBinaryWrappers/L_BFGS_B_jll.jl but it is not clear to me how you generated the .so binary.
The BB script: https://github.com/JuliaPackaging/Yggdrasil/tree/master/L/L_BFGS_B
You basically just need to use this: https://github.com/scipy/scipy/blob/master/scipy/optimize/lbfgsb_src/lbfgsb.f and recompile.
I guess we can apply a patch over the original version, but another thing I'm concerning is the license.
rht
commented
3 years ago You can view the licensing requirement of the 2011 version at https://github.com/scipy/scipy/blob/master/scipy/optimize/lbfgsb_src/README.
I additionally have noticed 2 changes to the API:
Gnimuc
commented
3 years ago That's the license of the original fortran code, I'm not sure whether it applies to the modified version or not. Anyway, feel free to open a PR and ignore the license issue at the moment.
rht
commented
3 years ago That 2011 version already includes the modification. I have inspected the source code of http://www.ece.northwestern.edu/~nocedal/lbfgsb.html.
It has the required c-jlm-jn comments.
c=========== L-BFGS-B (version 3.0. April 25, 2011 ===================
c
c This is a modified version of L-BFGS-B. Minor changes in the updated
c code appear preceded by a line comment as follows
c
c c-jlm-jn
cOK, I will make a PR.
Gnimuc
commented
3 years ago Feel free to ping me if you have any other questions :)
rht
commented
3 years ago Ah, I see. https://github.com/JuliaPackaging/Yggdrasil/blob/master/L/L_BFGS_B/build_tarballs.jl#L10 already uses the 3.0 version. I have also inspected the comment, also contains c-jlm-jn. But scipy.optimize's, lbfgsb.f is different since it has the extra maxls argument.
the first argument of setulb, n, is gone
This argument in scipy.optimize's lbfgsb.f is still there. It's just that scipy auto-computes it as len(x).
Gnimuc
commented
3 years ago You could download both the code from the official 3.0 tarball and the one from scipy, and then do a git diff to see the changes.
rht
commented
3 years ago OK, basically the main differences are
maxls becomes a parameter instead of being hardcoded to 20.itfile for iterate.dat is gone. The output is redirected to somewhere else.Seems like a useful change.
$ diff original.f lbfgsb.f 1,5d0
< c
< c L-BFGS-B is released under the “New BSD License” (aka “Modified BSD License”
< c or “3-clause license”)
< c Please read attached file License.txt
< c
48c43
< + task, iprint, csave, lsave, isave, dsave)
---
> + task, iprint, csave, lsave, isave, dsave, maxls)
52c47
< integer n, m, iprint,
---
> integer n, m, iprint, maxls,
143,144c138
< c When iprint > 0, the file iterate.dat will be created to
< c summarize the iteration.
---
> c
279c273
< + csave,lsave,isave(22),dsave)
---
> + csave,lsave,isave(22),dsave, maxls)
290c284
< + iprint, csave, lsave, isave, dsave)
---
> + iprint, csave, lsave, isave, dsave, maxls)
295c289,290
< + iwhere(n), indx2(n), isave(23)
---
> + iwhere(n), indx2(n), isave(23),
> + maxls
429,430c424
< c When iprint > 0, the file iterate.dat will be created to
< c summarize the iteration.
---
> c
486c480
< integer i,k,nintol,itfile,iback,nskip,
---
> integer i,k,nintol,iback,nskip,
494a489,490
> double precision dlamch
> external dlamch
499c495
< epsmch = epsilon(one)
---
> epsmch = 2 * dlamch('e')
547,553c543
< info = 0
<
< itfile = 8
< if (iprint .ge. 1) then
< c open a summary file 'iterate.dat'
< open (8, file = 'iterate.dat', status = 'unknown')
< endif
---
> info = 0
559c549
< call prn3lb(n,x,f,task,iprint,info,itfile,
---
> call prn3lb(n,x,f,task,iprint,info,
566c556
< call prn1lb(n,m,l,u,x,iprint,itfile,epsmch)
---
> call prn1lb(n,m,l,u,x,iprint,epsmch)
583d572
< itfile = isave(3)
649d637
< write (itfile,1003) iter,nfgv,sbgnrm,f
684c672
< if(iprint .ge. 1) write (6, 1005)
---
> if(iprint .ge. 1) write (0, 1005)
732c720
< if(iprint .ge. 1) write (6, 1006)
---
> if(iprint .ge. 1) write (0, 1006)
758c746
< if(iprint .ge. 1) write (6, 1005)
---
> if(iprint .ge. 1) write (0, 1005)
789,790c777,778
< + boxed,cnstnd,csave,isave(22),dsave(17))
< if (info .ne. 0 .or. iback .ge. 20) then
---
> + boxed,cnstnd,csave,isave(22),dsave(17), iprint)
> if (info .ne. 0 .or. iback .ge. maxls) then
809c797
< if(iprint .ge. 1) write (6, 1008)
---
> if(iprint .ge. 1) write (0, 1008)
837c825
< call prn2lb(n,x,f,g,iprint,itfile,iter,nfgv,nact,
---
> call prn2lb(n,x,f,g,iprint,iter,nfgv,nact,
907c895
< if(iprint .ge. 1) write (6, 1007)
---
> if(iprint .ge. 1) write (0, 1007)
930c918
< call prn3lb(n,x,f,task,iprint,info,itfile,
---
> call prn3lb(n,x,f,task,iprint,info,
943,944c931
< isave(1) = nintol
< isave(3) = itfile
---
> isave(1) = nintol
1090c1077
< if (prjctd) write (6,*)
---
> if (prjctd) write (0,*)
1093c1080
< + write (6,*) 'This problem is unconstrained.'
---
> + write (0,*) 'This problem is unconstrained.'
1364,1365c1351
< c When iprint > 0, the file iterate.dat will be created to
< c summarize the iteration.
---
> c
2456c2442
< + isave, dsave)
---
> + isave, dsave, iprint)
2461c2447
< + nbd(n), isave(2)
---
> + nbd(n), isave(2), iprint
2472a2459,2462
> c Be mindful that the dcsrch subroutine being called is a copy in
> c this file (lbfgsb.f) and NOT in the Minpack2 copy distributed
> c by scipy.
> c
2553c2543,2545
< write(6,*)' ascent direction in projection gd = ', gd
---
> if (iprint .ge. 0) then
> write(0,*)' ascent direction in projection gd = ', gd
> endif
2569a2562
> c take step and prevent rounding error beyond bound
2571a2565,2566
> if (nbd(i).eq.1.or.nbd(i).eq.2) x(i) = max(x(i), l(i))
> if (nbd(i).eq.2.or.nbd(i).eq.3) x(i) = min(x(i), u(i))
2671c2666
< subroutine prn1lb(n, m, l, u, x, iprint, itfile, epsmch)
---
> subroutine prn1lb(n, m, l, u, x, iprint, epsmch)
2673c2668
< integer n, m, iprint, itfile
---
> integer n, m, iprint
2703,2705d2697
< write (itfile,2001) epsmch
< write (itfile,*)'N = ',n,' M = ',m
< write (itfile,9001)
2742c2734
< subroutine prn2lb(n, x, f, g, iprint, itfile, iter, nfgv, nact,
---
> subroutine prn2lb(n, x, f, g, iprint, iter, nfgv, nact,
2746c2738
< integer n, iprint, itfile, iter, nfgv, nact, nseg,
---
> integer n, iprint, iter, nfgv, nact, nseg,
2796,2797c2788
< if (iprint .ge. 1) write (itfile,3001)
< + iter,nfgv,nseg,nact,word,iback,stp,xstep,sbgnrm,f
---
>
2810c2801
< subroutine prn3lb(n, x, f, task, iprint, info, itfile,
---
> subroutine prn3lb(n, x, f, task, iprint, info,
2817c2808
< integer n, iprint, info, itfile, iter, nfgv, nintol,
---
> integer n, iprint, info, iter, nfgv, nintol,
2860,2865c2851,2856
< if (info .eq. -1) write (6,9011)
< if (info .eq. -2) write (6,9012)
< if (info .eq. -3) write (6,9013)
< if (info .eq. -4) write (6,9014)
< if (info .eq. -5) write (6,9015)
< if (info .eq. -6) write (6,*)' Input nbd(',k,') is invalid.'
---
> if (info .eq. -1) write (0,9011)
> if (info .eq. -2) write (0,9012)
> if (info .eq. -3) write (0,9013)
> if (info .eq. -4) write (0,9014)
> if (info .eq. -5) write (0,9015)
> if (info .eq. -6) write (0,*)' Input nbd(',k,') is invalid.'
2868,2888c2859,2860
< if (info .eq. -8) write (6,9018)
< if (info .eq. -9) write (6,9019)
< endif
< if (iprint .ge. 1) write (6,3007) cachyt,sbtime,lnscht
< write (6,3008) time
< if (iprint .ge. 1) then
< if (info .eq. -4 .or. info .eq. -9) then
< write (itfile,3002)
< + iter,nfgv,nseg,nact,word,iback,stp,xstep
< endif
< write (itfile,3009) task
< if (info .ne. 0) then
< if (info .eq. -1) write (itfile,9011)
< if (info .eq. -2) write (itfile,9012)
< if (info .eq. -3) write (itfile,9013)
< if (info .eq. -4) write (itfile,9014)
< if (info .eq. -5) write (itfile,9015)
< if (info .eq. -8) write (itfile,9018)
< if (info .eq. -9) write (itfile,9019)
< endif
< write (itfile,3008) time
---
> if (info .eq. -8) write (0,9018)
> if (info .eq. -9) write (0,9019)
2931,2932c2903,2905
< +' Line search cannot locate an adequate point after 20 function',/
< +,' and gradient evaluations. Previous x, f and g restored.',/,
---
> +' Line search cannot locate an adequate point after MAXLS',/
> +,' function and gradient evaluations.',/
> +,' Previous x, f and g restored.',/,
2974a2948,2952
> if (gi.ne.gi) then
> c NaN value in gradient: propagate it
> sbgnrm = gi
> return
> endif
3147,3148c3125
< c When iprint > 0, the file iterate.dat will be created to
< c summarize the iteration.
---
> c
3282,3283c3259,3262
< write(6,*) ' Positive dir derivative in projection '
< write(6,*) ' Using the backtracking step '
---
> if (iprint .ge. 0) then
> write(6,*) ' Positive dir derivative in projection '
> write(6,*) ' Using the backtracking step '
> endifLooks like this is the feature mentioned in #2
rht
commented
3 years ago OK, this issue is a duplicate then.
Gnimuc
commented
3 years ago I think we should close #2 and keep this issue open.
Thank you for wrapping the library. Such a lifesaver. Someone in discourse.julialang.org mentioned that the lbfgsb.f used in
scipy.optimizeis actually a 2011 modification of the original method, see: https://github.com/JuliaNLSolvers/Optim.jl/issues/927.There is https://github.com/JuliaBinaryWrappers/L_BFGS_B_jll.jl but it is not clear to me how you generated the .so binary.