SymbolicRegression.jl simplification issue

[MilesCranmer]

I was wondering if there is a way to turn off automatic simplification? Perhaps there is something like #340 set up, also discussed in #327? cc @shashi @ChrisRackauckas @AlCap23

For example:

julia> using SymbolicUtils
julia> @syms x
julia> x * x
x^2

This automatic simplification seems to have broken SymbolicRegression.jl's simplification algorithm in the current SymbolicUtils.jl versions.

For example, if a user provides operators (+, *, /, -), then it is impossible for me to convert the expression:

x * x + x * x

back and forth between SymbolicRegression.jl and SymbolicUtils.jl. I can only go SymbolicRegression.jl => SymbolicUtils.jl, but not backwards.

What I did in previous versions of SymbolicUtils.jl was run simplify() as usual (defining a set of valid simplification rules given the operators, adding ^ to their set), but then convert x ^ 2 => x * x at the very end. However, it seems like I can no longer do this as even writing x * x simplifies directly to a power. i.e., the algebra is no longer literal:

julia> operation(x * x)
^ (generic function with 82 methods)

I also tried this:

julia> SymbolicUtils.show_simplified[] = false
julia> x * x
x^2

but the issue still occurs.

Is there anything I can do to prevent auto-simplification to operators I do not manually specify?

Thanks, Miles

[MilesCranmer]

For example, this is SymbolicUtils.jl v0.6:

julia> SymbolicUtils.operation(x * x)
* (generic function with 444 methods)

[MasonProtter]

Unfortunately, SymbllicUtils.jl bakes in some rules at the constructor level. So if you don’t want those rules, your only option is to roll your own types that don’t do those.

[MilesCranmer]

Is there a way I can manually create SymbolicUtils.jl types, rather than calling these auto-simplifying constructors? This would be good for me.

[ChrisRackauckas]

@shashi how hard would it be to support a @variables x::NoSimplifyReal type?

[shashi]

Yeah I'm gonna add it soon -- have an issue here https://github.com/JuliaSymbolics/SymbolicUtils.jl/issues/340