Re: [julia-users] MathProgBase Conic problem specifying dual
Hi Kevin! thanks letting me know; and for the quick response, I will re-post on julia-opt; best, steve On Sat, Jul 9, 2016 at 2:29 PM, Kevin Squire wrote: > Hi, it's probably worth asking this question on the julia-opt list. > > Cheers, >Kevin > > > > On Saturday, July 9, 2016, steven Varga wrote: > >> >> Is it possible to specify the dual of an SOCP/SDP using MathProgBase >> interface? >> >> It is well documented how to specify primal problem, but don't see any >> pointers to formulate the dual. The dual formulation is motivated if A >> matrix is non-square having more variables in primal then constraints. >> >> Any help, pointers appreciated >> steve >> >> MathProgBase version: >> commit 2faa7b97af385f61ea2e829512b5fa1c1c2ceab1 >> Author: Miles Lubin >> Date: Thu Jun 2 15:30:42 2016 +0200 >> >> Julia Version 0.5.0-dev+1503 >> Commit 84fb0a4* (2016-05-27 03:12 UTC) >> Platform Info: >> System: Linux (x86_64-linux-gnu) >> CPU: Intel(R) Core(TM) i7-3667U CPU @ 2.00GHz >> WORD_SIZE: 64 >> BLAS: libmkl_rt >> LAPACK: libmkl_rt >> LIBM: libimf >> >>
[julia-users] MathProgBase Conic problem specifying dual
Is it possible to specify the dual of an SOCP/SDP using MathProgBase interface? It is well documented how to specify primal problem, but don't see any pointers to formulate the dual. The dual formulation is motivated if A matrix is non-square having more variables in primal then constraints. Any help, pointers appreciated steve MathProgBase version: commit 2faa7b97af385f61ea2e829512b5fa1c1c2ceab1 Author: Miles Lubin Date: Thu Jun 2 15:30:42 2016 +0200 Julia Version 0.5.0-dev+1503 Commit 84fb0a4* (2016-05-27 03:12 UTC) Platform Info: System: Linux (x86_64-linux-gnu) CPU: Intel(R) Core(TM) i7-3667U CPU @ 2.00GHz WORD_SIZE: 64 BLAS: libmkl_rt LAPACK: libmkl_rt LIBM: libimf
[julia-users] REPL: show( io::IO, x::Array{my_object,1} ) = ... overwrite
When overwriting Base.show for an Array of custom_type REPL eval() uses the old version of Base.show with only partial delegation/dispatch; which is cumbersome when working with large arrays of records in REPL. Is this the expected behaviour, and there is a different approach to use REPL + eval + show or some sort of a exception I happened to stumble upon? any pointer/help is appreciated. [I am aware when wrapping arrays again in an object this can be avoided, but that requires implementing/dispatching methods Array{}() already has ] short self contained minimal example attached thanks, steve --- CODE --- #!/usr/local/bin/julia # minimal working example to demonstrate that one is # unable to overwrite Base.show for Array{custom_type,1} when using REPL # # julia> versioninfo() # Julia Version 0.4.0-dev+2895 # Commit a37025e* (2015-01-25 09:20 UTC) #Platform Info: # System: Linux (x86_64-linux-gnu) # CPU: Intel(R) Core(TM) i7-3667U CPU @ 2.00GHz # WORD_SIZE: 64 # BLAS: libmkl_rt # LAPACK: libmkl_rt # LIBM: libimf # LLVM: libLLVM-3.3 # module repl_test import Base.show export show type A value::Int64 end show( io::IO, x::A ) = @printf(" A: %i\n", x.value ) show( io::IO, x::Array{A,1} ) = @printf(" A: many values\n" ) end x = repl_test.A( 5 ) show(x) # as advertised x # same as above # printout in REPL # A: 5 # A: many values y = Array(repl_test.A, 10) show(y) # still works as expected # julia> show(y) # A: many values ## like io::IO, x::Array{A,1} = ... didn't exist y # julia> y # 10-element Array{repl_test.A,1}: # #undef # ... omitted # #undef # The above code when evaluated in REPL uses built in Base.show for array and indirectly # calls overwritten Base.show # I find it very cumbersome and unfriendly when unable to pretty print large arrays of structs # imported from other systems. # Is there a better/working solution to use the REPL similarly to when interacting with GNU R? repl_show_test.jl Description: Binary data
[julia-users] matrix shape subview of a large size vector with julia 0.3
Hello, As a beginner Julia user I have the outlined problem and a peer review request if there is any implications of the proposed technique. The problem: To have a matrix{T} shape continuous sub-view of a vector{T} store the following code, using pointer to array, seem to work: myView{T,N}(a::Array{T}, dims::NTuple{N,Int}; start::Int64=1 ) = pointer_to_array(pointer(a, start), dims, false ) v = zeros(20) ## must not be garbage collected until A,B referenced; v is 100M ~ 10G; in memory #for i=0:large_number # scatter A = myView(v, (10,1), start=1) B = myView(v, (10,1), start=11) # IO heavy updates A[:,:] = A[:,:] + 1; B[:,:] = B[:,:] + 5 #end # implicit gather show( v ) ## Is this a correct Julian way to implement efficient large size shared data-store, where A,B preserve matrix properties [BLAS/LAPACK]? Are there any obvious impacts/alternatives to this approach that avoided my attention? (cost of contracting sub-view versus BLAS.copy data back during gather) thanks, steve
[julia-users] julia stable compile from julia-0.3.3_b24213b893.tar.gz
Hi, I compiled the latest git version of julia with MKL on linux Mint 17; most function (BLAS,LAPACK,...) worked however didn't stable for production use (Winston, Gladfly failed to load). Then turned to the stable release which produces the following error: double_conversion_wrapper.cpp:1:31: fatal error: double-conversion.h: No such file or directory #include "double-conversion.h" I tried the git then the tar version; the problem persisted; Can someone give me a hint what I am doing wrong? steve