ENH: Implement gaussian process

[yibeichan]

1. improve calculate_angle_matrix by adding more checks for the input types and the dimensions
2. update compute_single_shell_covariance_matrix by (a) removing k from inputs and (b) calling compute_spherical_function in the calculation

notes: if we remove k in the input for the single shell, we probably want to be consistent and do it for multiple shell too

[oesteban]

@yibeichan @jhlegarreta how much of this can be split into smaller chunks and merged into the PR's branch?

[jhlegarreta]

On my ToDo list. The angle matrix computation part has been answered in PR nipreps#200.

I will have a look at the other suggestions as time permits. Given that we decided to split the whole GP draft PR into smaller patch sets for easier review and integration (e.g. nipreps#198, nipreps#199, nipreps#200, and upcoming ones) I will need to cherry pick the remaining suggestions in her.

[jhlegarreta]

Went through Yibei's changes on the single-shell covariance matrix computation and comments on the multi-shell case:

• The former will be addressed in a separate PR, once PR hyperparameters optimization #2 gets split into smaller PRs to the main repository (computation of covariance functions and the kernel; then the rest into other PRs). It essentially changed the bvec computation in the covariance method to index the pre-computed bvec angles (i.e. call the method in ENH: Add gradient encoding direction angle computation utils nipreps/eddymotion#200), and a call to the compute_spherical_function, which we may eventually do at https://github.com/jhlegarreta/eddymotion/pull/2/files#diff-7401613f4c745123dea57ac1f74163b4e14f6ed2e0c82fa2e466835139743fdbR61.
• The comments about the multi-shell case are fine; we are aware of the fact that things need to be adjusted/refactored as we gain insight and put things in place.

So closing.

Thanks for the work @yibeichan.