gnimuc/lbfgsb.jl

Julia wrapper for L-BFGS-B Nonlinear Optimization Code

lbfgsb.jl's Issues

LBFGSB still prints messages even with `iprint < 0`

The underlying Fortran code still prints messages regardless of iprint, which clogs up stdout pretty quickly in my case. This could be fixed fairly quickly by adding an if statement to lines 3282-3283 in lbfgsb.f:

write(6,*) ' Positive dir derivative in projection '
write(6,*) ' Using the backtracking step '

If this package uses the C version, the fix is detailed here. I'm not sure how to do it myself since I am unfamiliar with the new dependency-building mechanism.

Use a modified version of L-BFGS-B which is already used in scipy.optimize?

Thank you for wrapping the library. Such a lifesaver.
Someone in discourse.julialang.org mentioned that the lbfgsb.f used in scipy.optimize is actually a 2011 modification of the original method, see: JuliaNLSolvers/Optim.jl#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.

Is it possible to trace the optimization history?

Is it possible to know the loss/parameters in each step (rather than each function call).

A function fg! to compute both function and gradient would be more suitable for this software

Since the function and gradient evaluations occur always simultaneously, a function fg! to compute both function and gradient would be more suitable for this software (or at least supported as an option). This is especially desirable if the function and gradient evaluations share some computations.

TagBot trigger issue

This issue is used to trigger TagBot; feel free to unsubscribe.

If you haven't already, you should update your TagBot.yml to include issue comment triggers.
Please see this post on Discourse for instructions and more details.

If you'd like for me to do this for you, comment TagBot fix on this issue.
I'll open a PR within a few hours, please be patient!

Control maximum number of line search steps?

Just like what scipy did here:

https://github.com/scipy/scipy/blob/10233db264d15fa26ba3e2026006db95e59fd2c0/scipy/optimize/lbfgsb/lbfgsb.f#L778

Occasional printing even when `iprint<0`

Sometimes lbfgsb will print info:

ascent direction in projection gd =    0.0000000000000000

even though I have set iprint to -1.

Error Running Example Problem

Hello,

I'm trying to use LBFGSB but get the following error when running the example Rosenbrock script. I'm using version 1.2.0 of Julia and the error message is as follows:

ERROR: LoadError: MethodError: no method matching (::L_BFGS_B)(::typeof(f), ::typeof(g!), ::Array{Float64,1}, ::Array{Float64,2}; m=5, factr=1.0e7, pgtol=1.0e-5, iprint=-1, maxfun=15000, maxiter=15000)
Closest candidates are:
L_BFGS_B(::Any, ::AbstractArray{T,1} where T, ::AbstractArray{T,2} where T; m, factr, pgtol, iprint, maxfun, maxiter) at /Users/lab_home/.julia/packages/LBFGSB/mGpSN/src/wrapper.jl:33
Stacktrace:

This is the same error I was getting when I attempted to write my own problem as well. Any feedback/suggestions that you can give are much appreciated.

Fortran runtime error: End of record on 32bit Windows

CI keeps failing on AppVeyor, but I don't have a 32 bit Windows machine for debugging this issue for now:
https://ci.appveyor.com/project/Gnimuc/lbfgsb-jl/branch/master/job/43iwl8nm498eg57q

Building LBFGSB using julia compiled with MKL fails

I've been successfully using LBFGSB on julia 1.0.2, but recently I tried to compile julia with MKL to compare performances (still julia 1.0.2). When I try to build LBFGSB, I get the following error (I just changed the username):

┌ Warning: Cannot make sense of autodetected libstdc++ ABI version ('3.4.0')
└ @ BinaryProvider ~/.julia/packages/BinaryProvider/4F5Hq/src/PlatformNames.jl:624
┌ Debug: Rejecting cache file /home/USERNAME/.julia/compiled/v1.0/BinaryProvider/ek6VZ.ji due to it containing an invalid cache header
└ @ Base loading.jl:1330
┌ Debug: Recompiling stale cache file /home/USERNAME/.julia/compiled/v1.0/BinaryProvider/ek6VZ.ji for BinaryProvider [b99e7846-7c00-51b0-8f62-c81ae34c0232]
└ @ Base loading.jl:1190
┌ Warning: platform_key() is deprecated, use platform_key_abi() from now on
│   caller = ip:0x0
└ @ Core :-1
ERROR: LoadError: LibraryProduct(nothing, ["liblbfgsb"], :liblbfgsb, "Prefix(/home/USERNAME/.julia/packages/LBFGSB/sLXbo/deps/usr)") is not satisfied, cannot generate deps.jl!
Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] #write_deps_file#152(::Bool, ::Function, ::String, ::Array{LibraryProduct,1}) at /home/USERNAME/.julia/packages/BinaryProvider/4F5Hq/src/Products.jl:414
 [3] (::getfield(BinaryProvider, Symbol("#kw##write_deps_file")))(::NamedTuple{(:verbose,),Tuple{Bool}}, ::typeof(write_deps_file), ::String, ::Array{LibraryProduct,1}) at ./none:0
 [4] top-level scope at none:0
 [5] include at ./boot.jl:317 [inlined]
 [6] include_relative(::Module, ::String) at ./loading.jl:1044
 [7] include(::Module, ::String) at ./sysimg.jl:29
 [8] include(::String) at ./client.jl:392
 [9] top-level scope at none:0
in expression starting at /home/USERNAME/.julia/packages/LBFGSB/sLXbo/deps/build.jl:46

I'm not sure to understand what is happening and if the compilation with MKL is really the issue or not, but the library is present and can be opened:

julia> Libdl.open("/home/USERNAME/.julia/packages/LBFGSB/sLXbo/deps/usr/lib/liblbfgsb-3.so")
IOStream(<file /home/USERNAME/.julia/packages/LBFGSB/sLXbo/deps/usr/lib/liblbfgsb-3.so>)

Any hint?

[EDIT] Sorry, posted too quickly, I used open and not dlopen

julia> Libdl.dlopen("/home/USERNAME/.julia/packages/LBFGSB/sLXbo/deps/usr/lib/liblbfgsb-3.so")
ERROR: could not load library "/home/USERNAME/.julia/packages/LBFGSB/sLXbo/deps/usr/lib/liblbfgsb-3.so"
libgfortran.so.4: Ne peut ouvrir le fichier d'objet partagé: Aucun fichier ou dossier de ce type

So I guess this is a more general problem, probably not related directly to LBFGSB or MKL but just to the fact that julia was compiled from source…