LotkaVolterra manifest file outdated

[jd-lara]

Looks like the LotkaVolterra example folder has an outdated and invalid Manifest file with an old reference to a master of DataDrivenDiffEq

[2445eb08] DataDrivenDiffEq v0.5.5 `https://github.com/SciML/DataDrivenDiffEq.jl.git#master`

It also seems to have a lot of packages that aren't part of the experiment and there is no Project.toml

[ChrisRackauckas]

@AlCap23 did you update the project/manifest with the last update?

[AlCap23]

Ah. Nope. Not right now. I have to fix some issues on the current versioning of DataDriven due to the SymbolicArray introduction and will update within the next few hrs.

[AlCap23]

Not everything, but the most important parts are working again.

I took the liberty to tag the current status as a release, since the optimization hyperparameters / sparse optimizers changed.

[HenriLaurie]

I think I have a related problem. According to the stacktrace, the error is thrown from the following line

nn_problem = ContinuousDataDrivenProblem(X̂, tsample, DX = Ŷ)

saying that collocate_data is given a type signature it has no method for:

MethodError: no method matching collocate_data(::Matrix{Float32}, ::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}}, ::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}}, ::InterpolationMethod{UnionAll})
Closest candidates are:
  collocate_data(::AbstractVector{T} where T, ::AbstractVector{T} where T, ::AbstractVector{T} where T, ::InterpolationMethod) at /home/henri/.julia/packages/DataDrivenDiffEq/Kd4Hy/src/utils/collocation.jl:216
  collocate_data(::Any, ::Any, ::Any) at /home/henri/.julia/packages/DataDrivenDiffEq/Kd4Hy/src/utils/collocation.jl:187
  collocate_data(::Any, ::Any, ::InterpolationMethod) at /home/henri/.julia/packages/DataDrivenDiffEq/Kd4Hy/src/utils/collocation.jl:214

You may remember last year I used this example for teaching, this year the same.

[AlCap23]

Hi! This is strange, I cannot reproduce this. Could you provide a more extensive minimal example?

If I just do

using DataDrivenDiffEq
using LinearAlgebra

X = randn(2,100)
t = 0.0:1.0:99
DX = randn(2,100)

prob = ContinuousDataDrivenProblem(X, t, DX = DX)

I don't get an error, as it should be. Maybe I need to check on the dispatch. Otherwise, you could also use

prob = ContinuousDataDrivenProblem(X, DX, t = t)

[HenriLaurie]

Well, unfortunately that will take time. I'll try to get the relevant inputs as data.

[HenriLaurie]

By the way I am a bit preoccupied, thanks for the help.

[HenriLaurie]

No idea why, but this is what happened:

I reproduced the error in a notebook with nothing but hudson_bay.jl code, except that the two matrices in the call were written to files and read back in.

I then did what you did above, in a new context, reading in the matrix data from the same files. No error.

Might the difference be the code loading?

I get the following mysterious message after the using statements in hudson_bay.jl:

 Activating environment at `~/WorkInProgress/ODEswithJulia2021/Week 3/Project.toml`
┌ Info: DataDrivenDiffEq : OccamNet is available.
└ @ DataDrivenDiffEq /home/henri/.julia/packages/DataDrivenDiffEq/Kd4Hy/src/DataDrivenDiffEq.jl:140

... no such message after using DelimitedFiles, LinearAlgebra, DataDrivenDiffEq in the minimal, non-erroring context.

[HenriLaurie]

I should add that replacing nn_problem = ContinuousDataDrivenProblem(X̂, tsample, DX = Ŷ) with nn_problem = ContinuousDataDrivenProblem(X̂, tsample, Ŷ) removes the error message.

[HenriLaurie]

Here, finally, a MWE:

using LinearAlgebra, DataDrivenDiffEq

tsample = 0.0:0.5:20.0
X = rand(Float32, 2, 41)
Y = rand(Float32, 2, 41)

nn_problem = ContinuousDataDrivenProblem(X, tsample, DX=Y)

gives the error

MethodError: no method matching collocate_data(::Matrix{Float32}, ::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}}, ::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}}, ::InterpolationMethod{UnionAll})

Sorry about the toothless mouthing of this problem!

[AlCap23]

No worries. I just pasted it into the REPL and was able to reproduce the error.

This is due to the type of tsample being Float64 while the data in X and Y is Float32. I do need to fix this in the collocation call ( promote type earlier than on the construction of the Problem).

Changing it into

nn_problem = ContinuousDataDrivenProblem(X, Float32.(tsample), DX=Y)

Works for me.

[HenriLaurie]

Super. I'll use that workaround in the meantime.

On Wed, 28 Jul 2021 at 07:32, JuliusMartensen @.***> wrote:

No worries. I just pasted it into the REPL and was able to reproduce the error.

This is due to the type of tsample being Float64 while the data in X and Y is Float32. I do need to fix this in the collocation call ( promote type earlier than on the construction of the Problem).

Changing it into

nn_problem = ContinuousDataDrivenProblem(X, Float32.(tsample), DX=Y)

Works for me.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/ChrisRackauckas/universal_differential_equations/issues/36#issuecomment-888022826, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFCBGB73ZYHVHWSH5FQPJA3TZ6I67ANCNFSM47ESZXIA .