[General] Configurable problems

[ocots]

Add the ability to configure or change the problem parameters. The problem can be a callable struct with default parameter values set at creation, which can be updated when the struct is called.

[ocots]

Current state of the code

For the moment to get a problem, for instance with 100 steps, we do:

using OptimalControlProblems

using OptimalControl
docp = beam(OptimalControlBackend(); N=100)
nlp = nlp_model(docp)

# or

using JuMP
nlp = beam(JuMPBackend(); N=100)

and then to solve it:

using NLPModelsIpopt
nlp_sol = NLPModelsIpopt.ipopt(nlp)

And to get metadata:

metadata[:beam]

Change the problem parameters

1. For the moment we cannot change the problem parameters. We could imagine to do:

docp = beam(OptimalControlBackend(); N=100, tf=2)

# or

nlp = beam(JuMPBackend(); N=100, tf=2)

but in that case we must insure there is no conflict between problem parameters names and CTDirect.direct_transcription (or JuMP.Model) keyword arguments. That is we must be sure that no keyword has parameters names.

2. Another possibility is:

docp = beam(OptimalControlBackend(); N=100, parameters=(tf=2, ))

# or

nlp = beam(JuMPBackend(); N=100, parameters=(tf=2, ))

and we update only parameters contained in the parameters argument.

3. Another possibility is:

docp = beam(OptimalControlBackend(); N=100)
docp(; tf=2)

# or

docp = beam(OptimalControlBackend(); N=100)
nlp = nlp_model(docp)
nlp(; tf=2)

# or

nlp = beam(JuMPBackend(); N=100)
nlp(; tf=2)

and we update either the docp of nlp.

4. Another possibility is:

docp = beam(OptimalControlBackend(); N=100, parameters=(tf=1, ))
docp(; tf=2)

# or

docp = beam(OptimalControlBackend(); N=100, parameters=(tf=1, ))
nlp = nlp_model(docp)
nlp(; tf=2)

# or

nlp = beam(JuMPBackend(); N=100, parameters=(tf=1, ))
nlp(; tf=2)

Metadata

1. Maybe, it would be better to have:

metadata(:beam)

2. Another possibility:

docp = beam(OptimalControlBackend())
metadata(docp)

# or

docp = beam(OptimalControlBackend())
nlp = nlp_model(docp)
metadata(nlp)

# or

nlp = beam(JuMPBackend())
metadata(nlp)

[ocots]

@PierreMartinon @jbcaillau In #135 (comment) above, what is your preferences for:

• "Change the problem parameters" section,
• "Metadata" section?

[ocots]

Clean metadata and remove what is useless.

name
minimise
final_time 
...

[jbcaillau]

@ocots i suggest to keep it simple, passing the metadata straight away as in

docp = beam(OptimalControlBackend(); N=100, tf=2)

# or

nlp = beam(JuMPBackend(); N=100, tf=2)

[ocots]

Actually, I chose possibility 2 for the parameters:

docp = beam(OptimalControlBackend(); N=100, parameters=(tf=2, ))

# or

nlp = beam(JuMPBackend(); N=100, parameters=(tf=2, ))

since otherwise it is difficult to avoid any conflict in the names.

For the metadata, I chose possibility 1.