Operator count inconsistency with JuMP

[jd-lara]

When I run sum(GAEv) the operator counter in JuMP stays in 0 since it performs all the operations using add_to_expression!. However, if I run sum(PAE) the operator counter is different than 0.

Some algebra method might be causing a call to a + operator instead of add_to_expression.

[jd-lara]

@joaquimg here is an example of the issue.

using JuMP
using Random
m = Model() 
I = 1:200
J = 1:200
vals = rand(length(I))
@variable(m, x[I, J] >= 0)
expression_container = JuMP.Containers.DenseAxisArray{JuMP.GenericAffExpr{Float64, VariableRef}}(undef, I)
for i in J
    expression_container[i] = sum(x[i,:]) - vals[i]
end
@constraint(m, con, sum(expression_container) >= 0);
m.operator_counter
0

using ParameterJuMP
mwp = ModelWithParams()
I = 1:200
J = 1:200
vals = rand(length(I))
param_vals = add_parameters(mwp, vals)
@variable(mwp, x[I, J] >= 0)
expression_container = JuMP.Containers.DenseAxisArray{ParameterJuMP.DoubleGenericAffExpr{Float64, VariableRef, ParameterRef}}(undef, I)
for i in J
    expression_container[i] = sum(x[i,:]) - param_vals[i]
end
@constraint(mwp, con, sum(expression_container) >= 0);
mwp.operator_counter
548

For larger models, the way the expression algebra is implemented leads to the following warning in JuMP.

Warning: The addition operator has been used on JuMP expressions a large number of times. This warning is safe to ignore but may indicate that model generation is slower than necessary. For performance reasons, you should not add expressions in a loop. Instead of x += y, use add_to_expression!(x,y) to modify x in place. If y is a single variable, you may also use add_to_expression!(x, coef, y) for x += coef*y.

[jd-lara]

bump @joaquimg and @blegat about this. Any hints on how to address it?

[blegat]

In sum(x[i,:]) - param_vals[i] you call Base.- instead of add_to_expression, isn't that what's causing the counter to increase ?

[jd-lara]

I think if that were the case the counter would be other than 0 in the JuMP example

[jd-lara]

I solved it by calling .destructive_add! when adding PAE + variable.

[blegat]

So it was not in the sum ?

[jd-lara]

Yeah the sums of PAE are not adding counts.

[joaquimg]

You are calling destructive_add explicitly?
Where that is called in the example?
Doesn't looks like there is any PAE +
variable.
The problem is GAEv + param?

[jd-lara]

After some exploration, I realized what actually fixed the problem was the JuMP update to 0.20.0 and not the use of destructive_add, my guess is that it was this PR jump-dev/JuMP.jl#1902.

This is the test code and an example of the final implementation I did in PowerSimulations.

using ParameterJuMP
using JuMP
mwp = ModelWithParams()
I = 1:200
J = 1:200
vals = rand(length(I))
param_vals = add_parameters(mwp, vals)
@variable(mwp, x[I, J] >= 0)
expression_container = JuMP.Containers.DenseAxisArray{ParameterJuMP.DoubleGenericAffExpr{Float64, VariableRef, ParameterRef}}(undef, I)
for i in J
    expression_container[i] = sum(x[i,:])
    add_to_expression!(expression_container[i], - param_vals[i])
end
@constraint(mwp, con, sum(expression_container) >= 0);
mwp.operator_counter

In JuMP 0.19.2 and ParameterJuMP v0.1.1 this code results in 348 operations and in JuMP 0.20.0 and ParameterJuMP v0.1.2 results in 0.

Sorry for the confusion, it wasn't clear what might have fixed the issue.