SciML
Accessible internals for sparse regression under SINDy
I have precalculated estimate of candidate_matrix and also Derivative matrix… How do i use sparse regression to solve for sparse coefficient matrix ?? solve on SINDy assumes basis in symbolic form… But i have basis as candidate_matrix stored in a csv file…
Can we make little more flexible internals to achieve this ?
With #371 you can use the optimizer directly on the data ( see the tests in /lib/DataDrivenSR/test ).
X, DX = data
res = opt(X, DX, [options = DataDrivenCommonOptions()])
Which returns a tuple of coefficients, thresholds and iterations used.
https://github.com/SciML/DataDrivenDiffEq.jl/blob/11b21b255cac81ab4e10e006aff1af76a17bb7f9/lib/DataDrivenSR/test/runtests.jl#L14
Is this Ya candidate_matrix or basis_matrix ??
Ah, sorry! DataDrivenSparse sweat_smile
Have to find my way through the sublibs
I cannot update to 1.0.0.. possibly something clashing with existing packages ?
(SciML) pkg> st
Status `~/SciML/Project.toml`
[c29ec348] AbstractDifferentiation v0.4.3
[052768ef] CUDA v3.12.0
[479239e8] Catalyst v12.3.1
⌃ [2445eb08] DataDrivenDiffEq v0.8.6
[6748aba7] DeepEquilibriumNetworks v0.2.3
[aae7a2af] DiffEqFlux v1.52.0
[0c46a032] DifferentialEquations v7.6.0
[587475ba] Flux v0.13.7
[f6369f11] ForwardDiff v0.10.33
[b2108857] Lux v0.4.34
[961ee093] ModelingToolkit v8.33.1
[429524aa] Optim v1.7.3
[3bd65402] Optimisers v0.2.10
[7f7a1694] Optimization v3.9.2
[1dea7af3] OrdinaryDiffEq v6.31.2
[91a5bcdd] Plots v1.36.2
[1ed8b502] SciMLSensitivity v7.11.0
[0c5d862f] Symbolics v4.13.0
[e88e6eb3] Zygote v0.6.49
[9a3f8284] Random
Info Packages marked with ⌃ have new versions available
@AlCap23 never mind.. I created newer enviornment for 1.0.0
What is DataDrivenSparse ? It is throwing following error
julia> using DataDrivenDiffEq
julia> using DataDrivenSparse
ERROR: ArgumentError: Package DataDrivenSparse not found in current path.
- Run `import Pkg; Pkg.add("DataDrivenSparse")` to install the DataDrivenSparse package.
Stacktrace:
[1] macro expansion
@ ./loading.jl:1163 [inlined]
[2] macro expansion
@ ./lock.jl:223 [inlined]
[3] require(into::Module, mod::Symbol)
@ Base ./loading.jl:1144
(DataDrivenDiffEq) pkg> add DataDrivenSparse
ERROR: The following package names could not be resolved:
* DataDrivenSparse (not found in project, manifest or registry)
It is registered as of this morning, but might take a while. Did you do ] registry update ?
It is registered as of this morning, but might take a while. Did you do ] registry update ?
Yea.. but DataDrivenSparse is still not found .
Otherwise you could just use a direct DataDrivenProblem with a unit basis ( or without any basis and a sparse regression ). This will now create a basis with n states where the n is the number of features.
The docs are not quite updated, but going to the deployment of documenter might help.
https://datadriven.sciml.ai/dev/
Actually, i have basis matrix \theta pre-evaluated in matrix of numbers(these came from simulation).. My question was about to use it directly, so as to minimize || X_dot - \theta*X ||.. just exactly what SINDY does ! Our usual way in DataDrivenDiffEq is to construct basis from X state variables(from closed form functions)
You can just use the algorithm than or solve the problem without a basis using a direct problem with prob = DirectDataDrivenProblem(theta,DX).