Lab 2: use cvx.sum_squares

[slinderman]

Make clear that students should use cvx.sum_squares in the deconvolve implementation. Otherwise CVX makes some weird decisions about which solver to use. Can we explicitly tell CVX to try ECOS first?

[slinderman]

@noahcowan1 I just saw this note to myself from 2023. Did you happen to play around with the solver and/or specification of the objective? Maybe we should make a note to students on Ed if sum_squares is the preferred method.

[noahcowan1]

ah yeah. most of the errors are solved by CLARABEL at this point, but replacing SCS with ECOS (and pip installing ECOS as it is no longer a required dependency with CVXPY after the introduction of CLARABEL) seemed to work best for me. Not a big issue as SCS still works as a fallback for CLARABEL, but SCS takes much longer than ECOS (interior point methods truly are kinda magical).

Should I make an ed post on 1 using sum_squares for the deconvolve and 2 installing ECOS and falling back to ecos if the default doesn't work?