I replicated the optimization in the basic example notebook basic_example.ipynb with a modified verson of @alanedelman's proof-of-concept using the JuMP package for non-linear optimization (see JuMP_model.ipynb). It reproduces the optimal trajectories exactly (by eye) with a substantial 100x speed-up! My basic steepest descent (with momentum) algorithm took 8.4 second versus 0.09 seconds using JuMP.
I replicated the optimization in the basic example notebook
basic_example.ipynb
with a modified verson of @alanedelman's proof-of-concept using the JuMP package for non-linear optimization (seeJuMP_model.ipynb
). It reproduces the optimal trajectories exactly (by eye) with a substantial 100x speed-up! My basic steepest descent (with momentum) algorithm took 8.4 second versus 0.09 seconds using JuMP.