[BUG]: Symbolic regression fails with a dimension mismatch

[sdwfrost]

What happened?

Describe the bug 🐞

Symbolic regression fails on a problem where Sindy and DMD work, when linking DataDrivenDiffEq with SymbolicRegression. I've also filed the issue on DataDrivenDiffEq (SciML/DataDrivenDiffEq.jl#523), but @ChrisRackauckas seems to think it's a SymbolicRegression issue.

Expected behavior

A data-driven solution should be found.

Minimal Reproducible Example 👇

# Load libraries
using OrdinaryDiffEq
using DataDrivenDiffEq
using DataDrivenSparse
using DataDrivenSR
using ModelingToolkit

# Define the SIR model
function sir_ode(u,p,t)
    (s,i,r) = u
    (β,γ) = p
    ds = -β*s*i
    di = β*s*i - γ*i
    dr = γ*i
    [ds,di,dr]
end

p = [0.5,0.25]
u0 = [0.99, 0.01, 0.0]
tspan = (0.0, 40.0)
δt = 1

solver = ExplicitRK()
sir_prob = ODEProblem(sir_ode, u0, tspan, p)
sir_sol = solve(sir_prob, solver, saveat = δt)

# Define the data-driven problem
dd_prob = DataDrivenProblem(sir_sol)

# Define the basis
@parameters t
@variables (u(t))[1:3]
Ψ = Basis([u; u[1]*u[2]], u, independent_variable = t)

## Sparse regression (works)
res_sparse = solve(dd_prob, Ψ, STLSQ(),digits=1)

## Symbolic regression
eqsearch_options = SymbolicRegression.Options(binary_operators = [+, *],
                                              loss = L1DistLoss(),
                                              verbosity = -1, progress = false, npop = 30,
                                              timeout_in_seconds = 60.0)
eqs = EQSearch(eq_options = eqsearch_options)
res_sr = solve(dd_prob, Ψ, eqs)

Version

[8254be44] SymbolicRegression v0.14.5

Operating System

macOS

Interface

Julia REPL

Relevant log output

ERROR: BoundsError: attempt to access 0-element Vector{Any} at index [1]
Stacktrace:
[1] getindex
@ ./essentials.jl:13 [inlined]

Extra Info

No response

[MilesCranmer]

Are you able to reproduce it in pure SymbolicRegression.jl? I'm not familiar with the DataDrivenDiffEq.jl syntax unfortunately.

Can you also show the full traceback?

[MilesCranmer]

Oh, I think the issue is:

[8254be44] SymbolicRegression v0.14.5

The current version of SymbolicRegression.jl is 1.5.0. The last 0. version was 0.24.5. So 0.14.5 is quite old. Can you figure out why it's not updating?

[sdwfrost]

I'm not sure why it didn't update. I ran from a fresh environment to get 1.5.0, but I still get the same error. The code spends some time running, so it may be post-processing the output from SymbolicRegression that's giving it problems, which would be a DataDrivenDiffEq issue:

ERROR: BoundsError: attempt to access 0-element Vector{Any} at index [1]
Stacktrace:
  [1] throw_boundserror(A::Vector{Any}, I::Tuple{Int64})
    @ Base ./essentials.jl:14
  [2] getindex
    @ ./essentials.jl:916 [inlined]
  [3] (::DataDrivenSR.var"#15#23"{Vector{Float64}, Vector{Any}})(i::Int64)
    @ DataDrivenSR ~/.julia/packages/DataDrivenSR/w4puU/src/DataDrivenSR.jl:170
  [4] iterate
    @ ./generator.jl:48 [inlined]
  [5] _collect(c::Base.OneTo{…}, itr::Base.Generator{…}, ::Base.EltypeUnknown, isz::Base.HasShape{…})
    @ Base ./array.jl:800
  [6] collect_similar(cont::Base.OneTo{Int64}, itr::Base.Generator{Base.OneTo{…}, DataDrivenSR.var"#15#23"{…}})
    @ Base ./array.jl:709
  [7] map(f::Function, A::Base.OneTo{Int64})
    @ Base ./abstractarray.jl:3371
  [8] convert_to_basis(paretofrontier::Vector{…}, prob::DataDrivenDiffEq.InternalDataDrivenProblem{…})
    @ DataDrivenSR ~/.julia/packages/DataDrivenSR/w4puU/src/DataDrivenSR.jl:169
  [9] DataDrivenSR.SRResult(prob::DataDrivenDiffEq.InternalDataDrivenProblem{…}, hof::Vector{…}, paretos::Vector{…})
    @ DataDrivenSR ~/.julia/packages/DataDrivenSR/w4puU/src/DataDrivenSR.jl:71
 [10] (::EQSearch)(ps::DataDrivenDiffEq.InternalDataDrivenProblem{…}, X::Matrix{…}, Y::Matrix{…})
    @ DataDrivenSR ~/.julia/packages/DataDrivenSR/w4puU/src/DataDrivenSR.jl:203
 [11] (::DataDrivenSR.var"#26#27"{DataDrivenDiffEq.InternalDataDrivenProblem{…}, EQSearch})(::Tuple{Matrix{…}, Matrix{…}})
    @ DataDrivenSR ~/.julia/packages/DataDrivenSR/w4puU/src/DataDrivenSR.jl:214
 [12] iterate
    @ ./generator.jl:48 [inlined]
 [13] collect(itr::Base.Generator{DataLoader{…}, DataDrivenSR.var"#26#27"{…}})
    @ Base ./array.jl:780
 [14] map
    @ ./abstractarray.jl:3399 [inlined]
 [15] solve!(ps::DataDrivenDiffEq.InternalDataDrivenProblem{…})
    @ DataDrivenSR ~/.julia/packages/DataDrivenSR/w4puU/src/DataDrivenSR.jl:213
 [16] solve(::DataDrivenProblem{Float64, false, DataDrivenDiffEq.Continuous}, ::Vararg{Any}; kwargs::@Kwargs{})
    @ CommonSolve ~/.julia/packages/CommonSolve/JfpfI/src/CommonSolve.jl:23
 [17] top-level scope

[MilesCranmer]

Yeah it looks like the entire traceback is inside that package rather than SymbolicRegression.jl. Not sure there's anything I can do from my end. Maybe try to bisect the specific SymbolicRegression.jl version where it starts failing like this? https://github.com/MilesCranmer/SymbolicRegression.jl/releases. I'm assuming there is a version where this at one point worked?

[sdwfrost]

I'm assuming that there was a working version...however, v0.14.5 is the one that gets installed due to compatibility with DataDrivenSR, so I have to fiddle a bit more to try out lower version numbers. Closing for now - it looks like DataDrivenSR needs some love.