Gradient ascent trajectory function

[charlesrocabert]

Finalize a first version of the gradient ascent function:

• [ ] Define correct stop criteria to reach the (global/local) optimum
• [ ] Take care of negativity, and other algorithmic issues (too small dt, ...)
• [ ] Display and save properly the trajectory, and the estimated optimum
• [ ] As a next step, write a proper code suitable for a Python package

[fumer100]

This should check, if the Gradient is "0" right? Or do I miss something?

[charlesrocabert]

Yes. However I am not sure of the behavior of this stop criteria. I think it is better to check for changes in $\mu$ directly

[fumer100]

Yes. However I am not sure of the behavior of this stop criteria. I think it is better to check for changes in μ directly

Ok , right now I have a min. gradient of "-9.64858268e-06" for model A, which is veery close to 0. But I still did it with the "GCC_f check". I will try it with the μ changes. But is this gradient value correct? For Model A :

GCC_f is [ 0.00000000e+00    , -9.64858268e-06],
f is     [1.                 , 0.81269166]
v is     [912.25600275       , 741.38284929].

[charlesrocabert]

I can't tell. You need to plot the trajectories through time. What is the value of $\mu$? How does it increase through time? Try to generate some plots with matplotlib

[fumer100]

I think this is ok? Or should the growthrate be waay higher?

[charlesrocabert]

Correct!

[fumer100]

Ok, I am relieved :).