New release of CTBase, v0.7.0

[ocots]

See what's new here.

Still TBD:

• Minimal tests
• Improve abstract possibilities
• Add Poisson brackets
• Improve the documentation and add coverage

[ocots]

@PierreMartinon @joseph-gergaud @jbcaillau @BaptisteCbl Check this out!

[jbcaillau]

🙏🏽 thanks @ocots for the release and the recap: excellent! Having a look at the last changes.

@PierreMartinon @joseph-gergaud Functional & abstract descriptions are more or less stable 🤞🏾now, we count on you to retrieve a direct solve (= our basic tool) on the new version with variables. Everything should be OK from nlp_constraints (tested, but not yet heavily tested) but post an issue if not!

Also need to test the last ADNLP release from our colleagues with sparse Jacobians & Hessians and check performance increase 🤞🏾🤞🏾.

[BaptisteCbl]

Hello, if I'm not mistaken, with this new version, CTProblems need to be updated as well as CTFlows (which currently require the old CTBase 0.6.0). I can do pull requests for these two.

[ocots]

Thanks!

CTFlows must be updated before CTProblems. The scalar vs vectorial usage has been removed and this has to be taken into account.

You can make a try but I know that CTFlows code is not so clear.

You can also continue on CTBase since there are things to continue as differential geometry.

[BaptisteCbl]

Ok thanks, I'll finish things I can in CTBase before looking at it.

[ocots]

Update v0.7.1.

[joseph-gergaud]

With Pierre, we have done the modifications of CTDirect (trapeze). We are waiting the CTProbelm update for testing all the problems and to merge the branch. Here are some remarks / questions

• Is it possible to have a tag :test in CTProblem which would contain all problems for our tests ?
• It is possible for the moment that a user uses a scalar for the state in the définition of the state equation, but a vector for the lagrangian cost or for the initial condition. Do you think that it's possible to check this ?
• In the definition of the problem, it is possible to redefine the cost, but if we redefine a constraint (with the same tag), it's not possible. Before, we have to remove the constraint.
• as a user, I think that it's better to define the final time as free and not to define it as a variable.

[jbcaillau]

Salut @joseph-gergaud

With Pierre, we have done the modifications of CTDirect (trapeze). We are waiting the CTProbelm update for testing all the problems and to merge the branch. Here are some remarks / questions

👍🏽 good news!

• Is it possible to have a tag :test in CTProblem which would contain all problems for our tests ?

@ocots ?

• It is possible for the moment that a user uses a scalar for the state in the définition of the state equation, but a vector for the lagrangian cost or for the initial condition. Do you think that it's possible to check this ?

The user could do that; though not so coherent, it is correct julia syntax so it has to be allowed.

• In the definition of the problem, it is possible to redefine the cost, but if we redefine a constraint (with the same tag), it's not possible. Before, we have to remove the constraint.

I have just changed this: variable (if any), time, state, control, dynamics and objective declaration must be unique. PR to come.

• as a user, I think that it's better to define the final time as free and not to define it as a variable.

Disagree! If it is free, it is a variable (of the optimisation). This point of view allows us to treat free initial and / or final time and other optimisation variables uniformly. What is the point?

[jbcaillau]

@joseph-gergaud check #88

[jbcaillau]

@joseph-gergaud @PierreMartinon Actually, there are quite a few other tests that can be done already, apart from CTProblem's, so it would be nice that you merge in the develop branch of CTDirect.

[jbcaillau]

@ocots I’m not completely sure of the « good » way to test in this case: there are two packages = repos, A.jl and B.jl. If I want to use not the current release of B (using B...) from my local version of A (that I test using pkg> activate), can pkg> dev towards the appropriate branch of B.jl...?

[jbcaillau]

EDIT: should be OK as described here and there

[jbcaillau]

@joseph-gergaud @PierreMartinon Two questions:

• have you used the last version of ADNLP with sparse Jacobians & Hessians?
• it is now possible to define Mayer, Lagrange, and also Bolza cost: have you included this?

[jbcaillau]

@joseph-gergaud @PierreMartinon Nice job in CTDirect.jl, working fine with variables!

@def ocp4 begin
    tf ∈ R, variable
    t ∈ [ 0, tf ], time
    x ∈ R², state
    u ∈ R, control
    -1 ≤ u(t) ≤ 1
    x(0) == [ 0, 0 ]
    x(tf) == [ 1, 0 ]
    0.1 ≤ tf ≤ 10
    ẋ(t) == [ x₂(t), u(t) ]
    ∫( 1 ) → min
end

sol4 = solve(ocp4, grid_size=100, print_level=5, tol=1e-12)

plot(sol4)

Still an issue with scalar / vector variable, check #42