Unclear error in active_yet_argminmax

[gdalle]

While trying to debug #387, I ran into this error. I know there are NaNs somewhere in my code, but I'm not sure where it goes wrong

ERROR: BoundsError: attempt to access 1-element Vector{Float64} at index [-1]
Stacktrace:
  [1] getindex
    @ ./essentials.jl:13 [inlined]
  [2] getindex
    @ ~/.julia/packages/FrankWolfe/e97ZO/src/active_set.jl:48 [inlined]
  [3] #active_set_argminmax#84
    @ ~/.julia/packages/FrankWolfe/e97ZO/src/active_set.jl:271 [inlined]
  [4] active_set_argminmax
    @ ~/.julia/packages/FrankWolfe/e97ZO/src/active_set.jl:255 [inlined]
  [5] afw_step(x::Vector{Float64}, gradient::Vector{Float64}, lmo::InferOptFrankWolfeExt.LinearMaximizationOracleWithKwargs{typeof(ranking), Base.Pairs{Symbol, Matrix{Float32}, Tuple{Symbol}, NamedTuple{(:instance,), Tuple{Matrix{Float32}}}}}, active_set::FrankWolfe.ActiveSet{Vector{Int64}, Float64, Vector{Float64}}, epsilon::Float64, d::Vector{Float64}; memory_mode::FrankWolfe.InplaceEmphasis)
    @ FrankWolfe ~/.julia/packages/FrankWolfe/e97ZO/src/afw.jl:443
  [6] afw_step
    @ ~/.julia/packages/FrankWolfe/e97ZO/src/afw.jl:442 [inlined]
  [7] away_frank_wolfe(f::DifferentiableFrankWolfe.var"#obj#12"{Vector{Float32}, InferOptFrankWolfeExt.var"#f#3"{typeof(half_square_norm)}}, grad!::DifferentiableFrankWolfe.var"#grad!#13"{Vector{Float32}, InferOptFrankWolfeExt.var"#f_grad1#4"{typeof(identity)}}, lmo::InferOptFrankWolfeExt.LinearMaximizationOracleWithKwargs{typeof(ranking), Base.Pairs{Symbol, Matrix{Float32}, Tuple{Symbol}, NamedTuple{(:instance,), Tuple{Matrix{Float32}}}}}, active_set::FrankWolfe.ActiveSet{Vector{Int64}, Float64, Vector{Float64}}; line_search::FrankWolfe.Adaptive{Float64, Int64}, lazy_tolerance::Float64, epsilon::Float64, away_steps::Bool, lazy::Bool, momentum::Nothing, max_iteration::Int64, print_iter::Int64, trajectory::Bool, verbose::Bool, memory_mode::FrankWolfe.InplaceEmphasis, gradient::Nothing, renorm_interval::Int64, callback::Nothing, traj_data::Vector{Any}, timeout::Float64, extra_vertex_storage::Nothing, add_dropped_vertices::Bool, use_extra_vertex_storage::Bool, linesearch_workspace::Nothing, recompute_last_vertex::Bool)
    @ FrankWolfe ~/.julia/packages/FrankWolfe/e97ZO/src/afw.jl:228
  [8] away_frank_wolfe
    @ ~/.julia/packages/FrankWolfe/e97ZO/src/afw.jl:72 [inlined]
  [9] #away_frank_wolfe#106
    @ ~/.julia/packages/FrankWolfe/e97ZO/src/afw.jl:41 [inlined]
 [10] away_frank_wolfe
    @ ~/.julia/packages/FrankWolfe/e97ZO/src/afw.jl:10 [inlined]

[matbesancon]

Do you have access to the function that was called / a MWE?

[gdalle]

Unfortunately the MWE is rather involved, trying to debug it right now. But I thought it would already be useful to know that an end user can encounter this error. I'd love for it to be a little more explicit

[matbesancon]

yes it shouldn't happen, I could suspect the NaNs

[matbesancon]

because: julia> NaN < Inf

[gdalle]

fixed