search

jump-dev / Pajarito.jl

A solver for mixed-integer convex optimization
Mozilla Public License 2.0
131 stars 22 forks source link

Gurobi fails on "no conic solver" tests for Exp+SOC #333

Closed chriscoey closed 7 years ago

mlubin commented 7 years ago

what's the failure?

chriscoey commented 7 years ago 
julia> MathProgBase.optimize!(m)

Iter. | Best feasible  | Best bound     | Rel. gap    | Time (s)   
    1 |           +Inf |  -1.817678e+01 |         Inf |   8.508e-02
    2 |           +Inf |  -1.808333e+01 |         Inf |   9.116e-02
    3 |           +Inf |  -1.804110e+01 |         Inf |   9.631e-02
    4 |           +Inf |  -1.802048e+01 |         Inf |   1.013e-01
    5 |           +Inf |  -1.801023e+01 |         Inf |   1.063e-01
    6 |           +Inf |  -1.800512e+01 |         Inf |   1.112e-01
    7 |           +Inf |  -1.800256e+01 |         Inf |   1.175e-01
WARNING: Gurobi reported infeasible or unbounded. Set InfUnbdInfo=1 for more specific status.

Iterative algorithm summary:
 - Status               =     Infeasible
 - Best feasible        =           +Inf
 - Best bound           =  -1.800256e+01
 - Relative opt. gap    =            NaN
 - Total time (s)       =       3.37e-01

that is with our mip solver test tolerances:

tol_int = 1e-8
tol_feas = 1e-7
tol_gap = 1e-7

but with gurobi defaults, it doesn't fail.

chriscoey commented 7 years ago

with InfUnbdInfo=1 it says infeasible

mlubin commented 7 years ago

You can dump the MPS file from the last iteration and see what other solvers do.

On Mar 27, 2017 22:25, "Chris C." notifications@github.com wrote:

julia> MathProgBase.optimize!(m)

Iter. | Best feasible | Best bound | Rel. gap | Time (s) 1 | +Inf | -1.817678e+01 | Inf | 8.508e-02 2 | +Inf | -1.808333e+01 | Inf | 9.116e-02 3 | +Inf | -1.804110e+01 | Inf | 9.631e-02 4 | +Inf | -1.802048e+01 | Inf | 1.013e-01 5 | +Inf | -1.801023e+01 | Inf | 1.063e-01 6 | +Inf | -1.800512e+01 | Inf | 1.112e-01 7 | +Inf | -1.800256e+01 | Inf | 1.175e-01 WARNING: Gurobi reported infeasible or unbounded. Set InfUnbdInfo=1 for more specific status.

Iterative algorithm summary:

  • Status = Infeasible
  • Best feasible = +Inf
  • Best bound = -1.800256e+01
  • Relative opt. gap = NaN
  • Total time (s) = 3.37e-01

that is with our mip solver test tolerances:

tol_int = 1e-8 tol_feas = 1e-7 tol_gap = 1e-7

but with gurobi defaults, it doesn't fail.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/mlubin/Pajarito.jl/issues/333#issuecomment-289643411, or mute the thread https://github.com/notifications/unsubscribe-auth/ABp0M6ibt2pUzho74OL7lRVEFte5FVN8ks5rqG-YgaJpZM4Mq9ke .

chriscoey commented 7 years ago

passes on other solvers. I'm not debugging Gurobi so I'll close.