e0404 / apmtoolbox Goto Github PK

I have provided all my developed code in the branch dev_constrainedProbOpt in a subfolder cOpt. We have to merge this, which specifically affects

• anatomy definition; we should write a function that automatically creates different types of anatomies (only target, target+OAR, etc) for 1D prototypes for range and setup uncertainties, respectively.
• optimization & objective/constraint handling; The biological implementation considers probabilistic least squares represented as single, vectorized objective function over the whole anatomy. The constrained opt branch treats them structure based (therefore this is connected with the previous point).
• We have to think through if the constrained optimization is directly applicable with the biological models.
• we need to think about fractionation in the biological models?

In general, I think this should go hand in hand with modularization (see #1 ) to provide step by step modularized and compatible functionalities for

• anatomy creation
• modality selection and steering setup
• objective / constraint definition
• nominal optimization
• probabilistic optimization
• plotting / output

Ideas for consequent example scripts

• probabilistic least squares for protons / carbons (physical / biological)
• physical proton dose, probabilistic optimization with 1,2,...,Inf fractions
• comparison nominal / probabilistic DVH constraint lateral / range
• ...

For demo purposes it is really not so nice that all computations (exp calc, std calc, and sampling) are performed more or less in one go in the APM_1D_lateral. This really makes it harder to outsiders to understand what is going on. I suggest to have cells that can be executed one by one to run the individual computations.