Variable scaling and "Maximum_Iterations_Exceeded"

[Rémy MARION]

Hello,

I'm a new user of Casadi in Python. Especially with Opti stack.

I'm playing with Casadi to make a modelisation of a runner. My modelisation works fine for 100m race and 1500m. When I want to increase the lenght of the race (for example for a Marathon : 42195m), I have the message "Maximum_Iterations_Exceeded". I tried to increase the maximum iteration by using :

opti.solver("ipopt", {"expand":True}, {"max_iter":5000})

I have the same message "Maximum_Iterations_Exceeded". The solver should not converge. I decided so to try 2 solutions :

• Variable scaling for may be optimize the reaction of the solver (thre is an example in the Casadi blog) :

X = repmat([45000;7;2500],1,N).*opti.variable(3,N+1)        # variables d'état
pos    = X[0,:]                                                               # position
speed  = X[1,:]                                                             # vitesse
energy  = X[2,:]                                                            # energie
f = 15*opti.variable(1,N)                                                 # force
T = opti.variable()                                                         # temps final de la course

I took the wording in the Matlab example but it doesn't work with Python. What is the right wording for Python? pos is 42195 max, speed is 7 max, energy is 2500 max and f is 15 max.

I have the message :

    X = repmat([45000;7;2500],1,N).*opti.variable(3,N+1)        # variables d'état
                     ^
SyntaxError: invalid syntax

• Debugging using Callback function for trying understand why the solver doen't converge:
  

I tried the following code for trying catch the evolution of the T variable  :

opti.solver("ipopt", {"expand":True}, {"max_iter":5000})
sol = opti.solve()                                                        # lancement du solver

plt.figure(1)
opti.callback(lambda i: plt.plot(opti.debug.value(T)))
plt.show()

But no graph, nothing.

Do you have ideas ?

Thanks by advance.

Rémy

[Joris Gillis]

Hi Remy,

".*" means element-wise multiplication in Matlab. The Python equivalent is "*".
The callback command should be put before the solve call.

You may need to look up some tutorial on interactive plotting in matplotlib to tweak it a little bit.

Best regards,
   Joris

[Rémy MARION]

Hi Joris,

The Variable Scaling is very powerful. The calculation time decrease a lot for a more efficient results. It is clearly a key point. Everybody needs to read your blog.

But sometimes, the code doesn't converge. It is OK for 100m, 400m, 800m, 1500m, 10km and semi-marathon but when I want to make a calculation for 42195m (marathon), the solver failed (infeasible problem). It is however the same code !!! Of course, for each case, I tweak finely the Variable Scaling.

I saw that there is a new function in Opti.stack Casadi 3.4 to catch the infeasible problems. I will play with it for trying understand.

Thanks for your kelp. Step by step, I'm learning Optimisation, Casadi and for sure, one day, my code will be setted.

Rémy