Broadcasting over multiple DenseAxisArrays

[davide-f]

While trying to set up the initial point for a problem, the following problem occur.
no method matching set_start_value(::JuMP.Containers.DenseAxisArray{VariableRef,1,Tuple{Array{String,1}},Tuple{Dict{String,Int64}}}, ::JuMP.Containers.DenseAxisArray{Float64,1,Tuple{Array{String,1}},Tuple{Dict{String,Int64}}})

In attach a sample of the code
report_error.txt

[odow]

See the Note in the start values section of the docs: https://jump.dev/JuMP.jl/v0.21.1/variables/#Start-values-1. One option is (note the . after set_start_value):

set_start_value.(all_variables(model), value.(all_variables(model)))

Unfortunately, broadcasting over multiple DenseAxisArrays isn't supported yet, so you can't go
set_start_value.(x, value.(x)).

using JuMP
S = [:a, :b]
model = Model()
@variable(model, x[S], start = 0.0)
set_start_value.(x, start_value.(x))
ERROR: Broadcast operations with multiple DenseAxisArrays are not yet supported.
Stacktrace:
 [1] error(::String) at ./error.jl:33
 [2] broadcasted(::JuMP.Containers.DenseAxisArrayBroadcastStyle, ::Function, ::JuMP.Containers.DenseAxisArray{VariableRef,1,Tuple{Array{Symbol,1}},Tuple{Dict{Symbol,Int64}}}, ::JuMP.Containers.DenseAxisArray{Float64,1,Tuple{Array{Symbol,1}},Tuple{Dict{Symbol,Int64}}}) at /Users/oscar/.julia/packages/JuMP/YXK4e/src/Containers/DenseAxisArray.jl:249
 [3] broadcasted(::Function, ::JuMP.Containers.DenseAxisArray{VariableRef,1,Tuple{Array{Symbol,1}},Tuple{Dict{Symbol,Int64}}}, ::JuMP.Containers.DenseAxisArray{Float64,1,Tuple{Array{Symbol,1}},Tuple{Dict{Symbol,Int64}}}) at ./broadcast.jl:1237
 [4] top-level scope at REPL[21]:1

As a work-around, you can go

using JuMP
S = [:a, :b]
model = Model()
@variable(model, x[S], start = 0.0)
for i in eachindex(x)
    set_start_value(x[i], start_value(x[i]))
end

[chkwon]

Can the error message improved in this case?

using JuMP
S = [:a, :b]
model = Model()
@variable(model, x[S], start = 0.0)
set_start_value.(x, [0.0 1.0])

If the error message for this case could mention broadcasting, it would be helpful.

ERROR: MethodError: no method matching UnitRange{Int64}(::Array{Symbol,1})
Closest candidates are:
  UnitRange{Int64}(::Any, ::Static.StaticInt) where T<:Real at /Users/chkwon/.julia/packages/Static/XMB22/src/static_implementation.jl:133
  UnitRange{Int64}(::Any, ::Any) where T<:Real at range.jl:280
  UnitRange{Int64}(::UnitRange{T}) where T<:Real at range.jl:918
  ...
Stacktrace:
 [1] axistype(::Array{Symbol,1}, ::Base.OneTo{Int64}) at ./broadcast.jl:511
 [2] _bcs1(::Array{Symbol,1}, ::Base.OneTo{Int64}) at ./broadcast.jl:501
 [3] _bcs at ./broadcast.jl:495 [inlined]
 [4] broadcast_shape at ./broadcast.jl:489 [inlined]
 [5] combine_axes at ./broadcast.jl:484 [inlined]
 [6] instantiate at ./broadcast.jl:266 [inlined]
 [7] materialize(::Base.Broadcast.Broadcasted{Base.Broadcast.Unknown,Nothing,typeof(set_start_value),Tuple{JuMP.Containers.DenseAxisArray{VariableRef,1,Tuple{Array{Symbol,1}},Tuple{Dict{Symbol,Int64}}},Array{Float64,2}}}) at ./broadcast.jl:837
 [8] top-level scope at REPL[70]:1

[odow]

This is a tricky one that we haven't really addressed: people want a DenseAxisArray to broadcast like an Array of the same dimension.

At the very least, it deserves a better error message.