Replies: 12 comments 23 replies
-
Dans un premier temps, on aimerait avoir le problème de Goddard et d'autres cas d'usage sous les différentes formes1 données par le diagramme ci-dessous.
stateDiagram-v2
OCP --> JuMP.jl
OCP --> ExaModels.jl
OCP --> OptimalControl.jl
OptimalControl.jl --> ADNLPModel
JuMP.jl --> Ipopt.jl
ExaModels.jl --> ExaModel
ADNLPModel --> AbstractNLPModel
ExaModel --> AbstractNLPModel
AbstractNLPModel --> NLPModelsIpopt.jl
AbstractNLPModel --> MadNLP.jl
NLPModelsIpopt.jl --> Ipopt.jl
Footnotes
|
Beta Was this translation helpful? Give feedback.
-
Voulons-nous : flowchart TB
B(OCP) --> OptimalControl.jl --> Packages
Types --> AbstractNLPModel
AbstractNLPModel --> Wrappers
subgraph Modelisation
direction TB
subgraph Packages
ADNLPModels.jl
ExaModels.jl
JuMP.jl
end
subgraph Types
ADNLPModel
ExaModel
NLPModelsJuMP
end
ADNLPModels.jl --> ADNLPModel
ExaModels.jl --> ExaModel
JuMP.jl --> NLPModelsJuMP
AbstractNLPModel
end
subgraph Resolution
subgraph Wrappers
NLPModelsIpopt.jl
NLPModelsKnitro.jl
end
subgraph Solvers
Ipopt.jl
KNITRO.jl
MadNLP.jl
end
NLPModelsIpopt.jl --> Ipopt.jl
NLPModelsKnitro.jl --> KNITRO.jl
end
AbstractNLPModel --> MadNLP.jl
JuMP.jl --> Solvers
|
Beta Was this translation helpful? Give feedback.
-
On remarque deux choses :
|
Beta Was this translation helpful? Give feedback.
-
@frapac @ocots @PierreMartinon @joseph-gergaud
|
Beta Was this translation helpful? Give feedback.
-
PS. Congrats @frapac and co-authors !!! |
Beta Was this translation helpful? Give feedback.
-
Ideas for an internship: In a first place it could be great to compare some pairs of (model, solve) for some case studies: Goddard... See the list of problems here. The list of pairs of models and solvers (to be completed, updated) :
Then, it could be great to define a process for benchmarking with online results and with the possibility for users to make their own benchmarks. |
Beta Was this translation helpful? Give feedback.
-
@ocots @PierreMartinon @joseph-gergaud check this post on JuMP and related things (thanks @tmigot for pointing to ct). Most notably, function tracing now available in JuMP: @tmigot @amontoison does this entail full AD support (though not sparse, I guess) for the expressions generated by overloading on JuMP abstract expression type? |
Beta Was this translation helpful? Give feedback.
-
Stage M1 2024 : Livrables :
Remarque : ADNLPModel est utilisé via OptimalControl.jl. Pour le moment OptimalControl.jl ne fournit que ADNLPModel + Ipopt. Perspectives / Bonus :
|
Beta Was this translation helpful? Give feedback.
-
@amontoison @tmigot rappel des infos sur
|
Beta Was this translation helpful? Give feedback.
-
@frapac For the benchmark we should have a look to: |
Beta Was this translation helpful? Give feedback.
-
AD benchmark : JuliaSmoothOptimizers/ADNLPModels.jl#241 |
Beta Was this translation helpful? Give feedback.
-
L'équipe de contrôle optimal, vous utilisez quoi comme backend pour le Hessian creux dans ADNLPModels.jl? |
Beta Was this translation helpful? Give feedback.
-
@ocots @frapac @joseph-gergaud @PierreMartinon
Parfait
Oui : OptimalControl.jl peut être simplement utilisé comme une interface de modélisation ; que ce soit avec la syntaxe "abstraite" 1 ou "fonctionnelle" 2, le pb de contrôle optimal est traduit
Volontiers pour dév. commun, par exemple en s'appuyant sur un stage. Calons un point ensemble pour préciser.
Footnotes
voir cet exemple ↩
voir cet autre exemple ↩
Beta Was this translation helpful? Give feedback.
All reactions