The scope of the project

[larsgeb]

Andrea

I guess now comes the difficult part: to make decisions about what to put in the code and what kind of problems to address. For instance, do we want to have the core routines in e.g. C++ and then wrap everything in Python or mostly Python, and what about parallelization? My doubt right know is: should I code a high-performance version of HMC tomography and then wrap that in Python or should I keep the two things separate? My idea is that this package should be able to solve decent-size problems and be easy to use. If it's only a toy code I think people would tend to lose interest, thinking it's just a fancy but useless strategy.

[larsgeb]

I think we should decouple the forward/adjoint modules as much as possible from the sampler code. If we ensure that the sampler itself works properly on both laptops and HPC machinery, users are free to play with the actual physics / inverse problem.

To determine if you have to rewrite the non-linear HMC-tomography code:

• Is it programmed in a compiled language, such that it requires effort from the user to make it run?
• Does this decently run in its current form on a laptop machine for relevant problems (sizeable 2D/possibly 3D)?

Optimally, it would be fast enough to be relevant, but easy enough to get it to work.