Support blmo type and passed test for Cube polytope #188
Conversation
| n = 3 | ||
| k = 2 | ||
|
|
||
| Random.seed!(5) | ||
| n = 4 | ||
| k = 3 |
| x, _, _ = Boscia.solve(f, grad!, lmo, verbose=true) | ||
| x, _, _ = Boscia.solve(f, grad!, lmo, verbose=true, lazy=false, variant=Boscia.DICG()) |
There was a problem hiding this comment.
It would be nice to have both variants here. Also, it can function as a nice example for using DICG.
|
|
||
| param_mu = 0.4 | ||
| param_lambda = 0.1 | ||
|
|
||
| linVector = CSV.read("./data/qp_linVector.csv", DataFrame; header = false) |> Matrix | ||
| lapMatrix = CSV.read("./data/qp_lapMatrix.csv", DataFrame; header = false) |> Matrix | ||
| rrMatrix = CSV.read("./data/qp_rrMatrix.csv", DataFrame; header = false) |> Matrix | ||
| A = lapMatrix + param_mu * rrMatrix | ||
| @show size(A) | ||
| b = param_lambda * linVector | ||
|
|
||
| function f(x) | ||
|
|
||
| return (1 / 2) * (transpose(x) * A * x) + sum(transpose(b) * x) | ||
| end | ||
|
|
||
| function grad!(storage, x) | ||
| @. storage = $(A * x) + b | ||
| end | ||
|
|
||
| return f, grad! |
| seed = rand(UInt64) | ||
| @show seed | ||
| Random.seed!(seed) |
There was a problem hiding this comment.
Please comment if not needed so the whole test suite runs with the same seed.
There was a problem hiding this comment.
I think this is a nice additional example in Boscia. Please redo it in the classical example format. It is enough if we have one Boscia run using DICG.
src/blmo_interface.jl
Outdated
There was a problem hiding this comment.
Have you added dummies for the other DICG methods also?
src/polytope_blmos.jl
Outdated
| # Otherwise, assgin one to the entry corresponding to the smallest entry of d. | ||
| # Same for the UnitSimplex and ProbabilitySimplex. | ||
| function bounded_dicg_maximum_step(sblmo::CubeSimpleBLMO, x, direction, lb, ub, int_vars; kwargs...) | ||
| return FrankWolfe.dicg_maximum_step(FrankWolfe.ZeroOneHypercube(), x, direction) |
There was a problem hiding this comment.
In our case, the hypercube does not necessarily correspond to the Zero One hypercube!
src/polytope_blmos.jl
Outdated
| end | ||
|
|
||
| function bounded_dicg_maximum_step(sblmo::ProbabilitySimplexSimpleBLMO, x, direction, lb, ub, int_vars; kwargs...) | ||
| return FrankWolfe.dicg_maximum_step(FrankWolfe.ProbabilitySimplexOracle{Float64}(), x, direction) |
There was a problem hiding this comment.
You have to give the Probability Simplex the correct right side. Aka N in our case.
src/polytope_blmos.jl
Outdated
|
|
||
|
|
||
| function bounded_dicg_maximum_step(sblmo::UnitSimplexSimpleBLMO, x, direction, lb, ub, int_vars; kwargs...) | ||
| return FrankWolfe.dicg_maximum_step(FrankWolfe.UnitSimplexOracle{Float64}(), x, direction) |
test/simple_test_ODWB.jl
Outdated
There was a problem hiding this comment.
Please remove this. I have written an example in the example folder.
|
In terms of formatting, you have a lot of large indents. Please only use one tab, e.g. |
Modified code in frank_wolfe_variants.jl, Boscia.jl, node.jl, polytope_blmos.jl, time_tracking_lmo.jl, managed_blmo.jl.
All changed or added code is wrapped by
#=========================#
......
#=========================#