ArgumentError: invalid index: "s1" of type String

Specific Domains Optimization (Mathematical)
, ,
1

Hi,

Here is a MWE of my problem that is returning the same error. I cannot seem to understand why.

using DataStructures
using JuMP
using TaylorSeries

S = ["s1","s2","s3","s4"]
N = [1:10]

price = OrderedDict(
"s1" => 0,
"s2" => 0,
"s3" => 0,
"s4" => 5
)

m = Model()

@variables m begin
  d[s in S, n in N] >= 0
end

g(d)= sum((price[s]*d[s,n]) for n in N for s in keys(price))

∫⬩dn(N::Taylor1) = integrate(p)

function Taylor(f, p0)
    
    p = copy(p0)
    pnew = p0 + ∫⬩dn(f(N))
    
    while (pnew - p) != 0
        p = pnew
        pnew = p0 + ∫⬩dn(f(N))
    end
    
    return p
        
end

degree = 3

p0 = Taylor1([1], degree)

SurrogateApproximation = Taylor(g, p0)

When the equation is on its own, it does not return that error but returns it when called into the function Taylor

sum((price[s]*d[s,n]) for n in N for s in keys(price))

Here is the full details of the error

ArgumentError: invalid index: “s1” of type String

Stacktrace:
[1] to_index(::String) at .\indices.jl:297
[2] to_index(::Array{UnitRange{Int64},1}, ::String) at .\indices.jl:274
[3] to_indices at .\indices.jl:325 [inlined]
[4] to_indices at .\indices.jl:321 [inlined]
[5] getindex at .\abstractarray.jl:1060 [inlined]
[6] (::var"#393#396"{UnitRange{Int64},Array{UnitRange{Int64},1}})(::String) at .\none:0
[7] MappingRF at .\reduce.jl:93 [inlined]
[8] _foldl_impl(::Base.MappingRF{var"#393#396"{UnitRange{Int64},Array{UnitRange{Int64},1}},Base.BottomRF{typeof(Base.add_sum)}}, ::Base._InitialValue, ::Base.KeySet{String,OrderedDict{String,Int64}}) at .\reduce.jl:58
[9] FlatteningRF at .\reduce.jl:119 [inlined]
[10] MappingRF at .\reduce.jl:93 [inlined]
[11] _foldl_impl(::Base.MappingRF{var"#394#395"{Array{UnitRange{Int64},1}},Base.FlatteningRF{Base.BottomRF{typeof(Base.add_sum)}}}, ::Base._InitialValue, ::Array{UnitRange{Int64},1}) at .\reduce.jl:58
[12] foldl_impl at .\reduce.jl:48 [inlined]
[13] mapfoldl_impl at .\reduce.jl:44 [inlined]
[14] #mapfoldl#204 at .\reduce.jl:160 [inlined]
[15] mapfoldl at .\reduce.jl:160 [inlined]
[16] #mapreduce#208 at .\reduce.jl:287 [inlined]
[17] mapreduce at .\reduce.jl:287 [inlined]
[18] sum at .\reduce.jl:494 [inlined]
[19] sum(::Base.Iterators.Flatten{Base.Generator{Array{UnitRange{Int64},1},var"#394#395"{Array{UnitRange{Int64},1}}}}) at .\reduce.jl:511
[20] h(::Array{UnitRange{Int64},1}) at .\In[29]:1
[21] Taylor(::typeof(h), ::Taylor1{Int64}) at .\In[21]:4
[22] top-level scope at In[31]:5
[23] include_string(::Function, ::Module, ::String, ::String) at .\loading.jl:1091

2

Could this be because the variable d has indices [s,n]?

3

Not sure if this is related to your problem, I did not test it, but I am almost sure you want N = 1:10 and not N = [1:10], can you test with this change?

4

It returns the exact same error.

5

It’s worth stepping back and asking “what are you trying to do”?

Your g function takes an argument d indexed over s and n. But you pass g to Taylor with a p0 that is some sort of 3-degree expansion, and isn’t indexed over s and n.

6

Thank you for your note.

I am trying to build a taylor expansion of g. Is there a problem with my formulation considering what I am trying to do?

7

I am trying to build a taylor expansion of g.

To rephrase, why are you trying to do this? What is the bigger goal?

Is there a problem with my formulation considering what I am trying to do?

Yes. d should be a 2-dimensional DenseAxisArray, indexed by Strings and Ints with 40 elements. But you are trying to do a univariate Taylor series expansion.

8

The taylor expansion will act as a surrogate/approximation function that I will use in further steps.

9

Hmmm… I think I tried to used TalyorN for multivariate but it returned a similar error or rather my formulation is incorrect.

10

The taylor expansion will act as a surrogate/approximation function that I will use in further steps.

Yes. But why a surrogate, and why Taylor series?

  • Your function g is linear, you don’t need to approximate. Have you tried just passing the full problem to JuMP?
  • Why is JuMP in your MWE? It isn’t used.
  • Is there a bug in your Taylor function? The while loop always exits after the first iteration, because pnew never changes.
  • TaylorSeries is for low-dimensional problems. Think one, or several. Your d matrix has 40 elements. Are you trying to fit a surrogate function to a 40-dimensional surface? You may need to reconsider your approach.
1 Like
11

yes I have and it initially gave me problems. because it was a JuMP @objective which gives a GenericAffExpr. I will try it again this way

12

The variable d is a one of the variables in the full model. Hence the use of JuMP