New TMC modes: sequential, matrix and diagonal

[SteSeg]

This PR is a follow up of #79, Introducing different modes for running the TMC uncertainty quantification.

TMC Modes

Modes available are sequential, matrix and diagonal.

sequential is the one already available implemented with #79. namely, performs one perturbation at a time (with all its realizations), for a total of perturbations-times-realizations (r*p) models/simulations/tmc-iterations. Useful for identifying the effect of every single perturbation type to the total uncertainty.

matrix mode builds a matrix of perturbations and realizations. So, for every new realization of a single perturbation, the TMCManager runs all the realizations of all the other perturbations. Such matrix of perturbations-realizations would hence have as many dimensions as the number of perturbations (p) and every dimension is as long as the number of realizations (r), ending up with a total number of models/simulations/tmc-iterations as r^p. Useful to cover the whole perturbation-realization "phase-space", but often computationally and memory prohibitive.

diagonal mode takes the matrix of the matrix mode and runs only the models on the diagonal. On the diagonal all the perturbations get a new realization (hence a new value of the perturbed parameter) simultaneously achieving, arguably, a faster convergence to the total uncertainty. Useful to get faster to total uncertainty with a TMC method.

Usage

The usage is similar, the user just needs to specify the mode to run with the TMCManager.run method:

# Instantiate tmc manager
tmc_manager = ofb.uq.TMCManager(base_model=model, perturbations=[perturb_water_density], realizations=100, rng=np.random.default_rng(42))
# Run tmc
tmc_manager.run(mode="diagonal")