Limitations of UQPy with respect to memory, surrogate evaluations, etc. for PCE

[oarcelus]

Has anyone tried to benchmark UQPy against other python packages for PCE analysis such as Chaospy? I have found few bottlenecks on the latter library, especially regarding the efficiency of generation of polynomial expansions and evaluation of surrogates. For a full-order polynomial expansion with 15 unkowns and 6th order polynomials, I am already at 55k expansion elements. My analysis starts to take a huge amount of memory for 20k evaluations (with 10 targets) and long times.

Also is it possible to extract the sensing matrices as numpy arrays and operate on them on our own? For example evaluate LOO error with our own logic, or use different regression methods.