Docs for NeuralSurrogate

[jdossgollin]

Three issues with the NeuralSurrogate docs

First, loading the recommended packages yields a warning

julia> using Plots

julia> using Surrogates

julia> using Flux
┌ Warning: Package Surrogates does not have Flux in its dependencies:
│ - If you have Surrogates checked out for development and have
│   added Flux as a dependency but haven't updated your primary
│   environment's manifest file, try `Pkg.resolve()`.
│ - Otherwise you may need to report an issue with Surrogates
└ Loading Flux into Surrogates from project dependency, future warnings for Surrogates are suppressed.

This seems OK but it would be helpful to have a heads up for users.

Next, an error.
Following along, there is no lb or ub:

julia> neural = NeuralSurrogate(x, y, lb, ub, model = model1, n_echos = 10)
ERROR: UndefVarError: lb not defined
Stacktrace:
 [1] top-level scope

Next if I make up some numbers, there is a typo

lb, ub = -10, 10
surrogate_optimize(schaffer, SRBF(), lower_bound, upper_bound, neura, SobolSample(), maxiters=20, num_new_samples=10)
ERROR: UndefVarError: neura not defined
Stacktrace:
 [1] top-level scope

but even if I fix the typo I get an issue

neural = NeuralSurrogate(x, y, lb, ub, model = model1, n_echos = 10)
surrogate_optimize(schaffer, SRBF(), lower_bound, upper_bound, neural, SobolSample(), maxiters=20, num_new_samples=10)
ERROR: InexactError: Int64(3.8564731213645262)
Stacktrace:
 [1] Int64
   @ ./float.jl:723 [inlined]
 [2] convert
   @ ./number.jl:7 [inlined]
 [3] setindex!(A::Vector{Int64}, x::Float64, i1::Int64)
   @ Base ./array.jl:839
 [4] surrogate_optimize(obj::typeof(schaffer), ::SRBF, lb::Vector{Float64}, ub::Vector{Float64}, surr::NeuralSurrogate{UnitRange{Int64}, UnitRange{Int64}, Chain{Tuple{Dense{typeof(σ), Matrix{Float32}, Vector{Float32}}, Dense{typeof(σ), Matrix{Float32}, Vector{Float32}}, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}, Surrogates.var"#237#241"{Chain{Tuple{Dense{typeof(σ), Matrix{Float32}, Vector{Float32}}, Dense{typeof(σ), Matrix{Float32}, Vector{Float32}}, Dense{typeof(identity), Matrix{Float32}, Vector{Float32}}}}}, Descent, Zygote.Params, Int64, Int64, Int64}, sample_type::SobolSample; maxiters::Int64, num_new_samples::Int64)
   @ Surrogates ~/.julia/packages/Surrogates/lWGHe/src/Optimization.jl:117
 [5] top-level scope

[ChrisRackauckas]

Interesting, the test seems to be fine though? https://github.com/SciML/Surrogates.jl/blob/master/test/neuralSurrogate.jl