DAE optimizers added to OptimizationODE #932
Merged
+490
−71
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Comment on lines
66
to
81
| function solve_constrained_root(cache, u0, p) | ||
| n = length(u0) | ||
| cons_vals = cache.f.cons(u0, p) | ||
| m = length(cons_vals) | ||
| function resid!(res, u) | ||
| temp = similar(u) | ||
| f_mass!(temp, u, p, 0.0) | ||
| res .= temp | ||
| end | ||
| u0_ext = vcat(u0, zeros(m)) | ||
| prob_nl = NonlinearProblem(resid!, u0_ext, p) | ||
| sol_nl = solve(prob_nl, Newton(); tol = 1e-8, maxiters = 100000, | ||
| callback = cache.callback, progress = get(cache.solver_args, :progress, false)) | ||
| u_ext = sol_nl.u | ||
| return u_ext[1:n], sol_nl.retcode | ||
| end |
Comment on lines
113
to
62
| function finite_difference_jacobian(f, x; ϵ = 1e-8) | ||
| n = length(x) | ||
| fx = f(x) | ||
| if fx === nothing | ||
| return zeros(eltype(x), 0, n) | ||
| elseif isa(fx, Number) | ||
| J = zeros(eltype(fx), 1, n) | ||
| for j in 1:n | ||
| xj = copy(x) | ||
| xj[j] += ϵ | ||
| diff = f(xj) | ||
| if diff === nothing | ||
| diffval = zero(eltype(fx)) | ||
| else | ||
| diffval = diff - fx | ||
| end | ||
| J[1, j] = diffval / ϵ | ||
| end | ||
| return J | ||
| else | ||
| m = length(fx) | ||
| J = zeros(eltype(fx), m, n) | ||
| for j in 1:n | ||
| xj = copy(x) | ||
| xj[j] += ϵ | ||
| fxj = f(xj) | ||
| if fxj === nothing | ||
| @inbounds for i in 1:m | ||
| J[i, j] = -fx[i] / ϵ | ||
| end | ||
| else | ||
| @inbounds for i in 1:m | ||
| J[i, j] = (fxj[i] - fx[i]) / ϵ | ||
| end | ||
| end | ||
| end | ||
| return J | ||
| end | ||
| end |
| if solver_method == :mass_matrix | ||
| return solve_dae_mass_matrix(cache, dt, maxit, u0, p) | ||
| else | ||
| return solve_dae_indexing(cache, dt, maxit, u0, p, differential_vars) |
There was a problem hiding this comment.
I assume you mean implicit?
Comment on lines
166
to
167
| solver_method = get_solver_type(cache.opt) | ||
| if solver_method == :mass_matrix |
There was a problem hiding this comment.
The actual check is alg isa DAEAlgorithm which means implicit, otherwise mass matrix
| end | ||
|
|
||
| if m == 0 | ||
| optf = ODEFunction(f_mass!, mass_matrix = I(n)) |
There was a problem hiding this comment.
Suggested change
| optf = ODEFunction(f_mass!, mass_matrix = I(n)) | |
| optf = ODEFunction(f_mass!) |
| if m == 0 | ||
| optf = ODEFunction(f_mass!, mass_matrix = I(n)) | ||
| prob = ODEProblem(optf, u0, (0.0, 1.0), p) | ||
| return solve(prob, HighOrderDescent(); dt=dt, maxiters=maxit) |
There was a problem hiding this comment.
why is the solver being specified?
There was a problem hiding this comment.
when falling back to ode, I thought it better to use the best method.
There was a problem hiding this comment.
That's not necessarily the best method.
| n = length(u0) | ||
| m = length(cons_vals) | ||
| u0_extended = vcat(u0, zeros(m)) | ||
| M = zeros(n + m, n + m) |
There was a problem hiding this comment.
Just make M a diagonal matrix
Comment on lines
246
to
168
| function f_mass!(du, u, p_, t) | ||
| x = @view u[1:n] | ||
| λ = @view u[n+1:end] | ||
| grad_f = similar(x) | ||
| if cache.f.grad !== nothing | ||
| cache.f.grad(grad_f, x, p_) | ||
| else | ||
| grad_f .= ForwardDiff.gradient(z -> cache.f.f(z, p_), x) | ||
| end | ||
| J = Matrix{eltype(x)}(undef, m, n) | ||
| if cache.f.cons_j !== nothing | ||
| cache.f.cons_j(J, x) | ||
| else | ||
| J .= finite_difference_jacobian(z -> cache.f.cons(z, p_), x) | ||
| end | ||
| @. du[1:n] = -grad_f - (J' * λ) | ||
| consv = cache.f.cons(x, p_) | ||
| if consv === nothing | ||
| fill!(du[n+1:end], zero(eltype(x))) | ||
| else | ||
| if isa(consv, Number) | ||
| @assert m == 1 | ||
| du[n+1] = consv | ||
| else | ||
| @assert length(consv) == m | ||
| @. du[n+1:end] = consv | ||
| end | ||
| end | ||
| return nothing | ||
| end |
There was a problem hiding this comment.
Unnecessary, it's just 0
Comment on lines
330
to
225
| if cache.f.cons_j !== nothing | ||
| cache.f.cons_j(J, x) | ||
| else | ||
| J .= finite_difference_jacobian(z -> cache.f.cons(z,p_), x) | ||
| end | ||
| @. res[1:n] = du_x + grad_f + J' * λ |
There was a problem hiding this comment.
It's just zero = cons for the other terms.
Comment on lines
67
to
71
| if opt.solver isa Union{Rodas5, RadauIIA5, ImplicitEuler, Trapezoid} | ||
| return :mass_matrix | ||
| else | ||
| return :indexing | ||
| end |
There was a problem hiding this comment.
Use opt.solver isa DAEAlgorithm instead
There was a problem hiding this comment.
This has been removed in the latest commits.
Docs for first draft of OptimizationODE.jl
Co-authored-by: Christopher Rackauckas <[email protected]>
Method descriptions for the solvers added.
Updated docs.
updated docs
Removed redundant functions, used proper jacobian.
Code changes based on review.
Jacobian fix for tests.
Removed hard coded alg call.
Linear Algebra dependency added.
added package dependencies
Removed redundant tests.
Mismatch on problem and algorithm fixed.
Correcting function definition.
ChrisRackauckas
force-pushed
the
daenew
branch
from
e13cbca to
980e11d
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Checklist
contributor guidelines, in particular the SciML Style Guide and
COLPRAC.
Additional context
Add any other context about the problem here.