Improve computational speed with wls (and non-gaussian?)

[MartaVanin]

just a reminder and a place to put new ideas if they pop-up

• try different rescalers
• try relaxed wls
• try stricter variable bounds
• compare different (exact) formulations to see what's faster
• talk to people (Frederik, ...)

[timmyfaraday]

You forgot:

• testing whether the bounds on rho are an issue

[MartaVanin]

For computational time? Because I added a big M and now the logpdf works: it converges to the global minimum, when it converges 😝

From: Tom Van Acker notifications@github.com Sent: Tuesday, 8 September 2020 16:35 To: timmyfaraday/PowerModelsDSSE.jl PowerModelsDSSE.jl@noreply.github.com Cc: MartaVanin marta.vanin@esat.kuleuven.be; Author author@noreply.github.com Subject: Re: [timmyfaraday/PowerModelsDSSE.jl] Improve computational speed with wls (and non-gaussian?) (#19)

You forgot:

• testing whether the bounds on rho are an issue

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/timmyfaraday/PowerModelsDSSE.jl/issues/19#issuecomment-688919240, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AFCVFYPD33XANRQ3C7PFGDLSEY6JHANCNFSM4RABPT3A. [ { "@context": "http://schema.org", "@type": "EmailMessage", "potentialAction": { "@type": "ViewAction", "target": "https://github.com/timmyfaraday/PowerModelsDSSE.jl/issues/19#issuecomment-688919240", "url": "https://github.com/timmyfaraday/PowerModelsDSSE.jl/issues/19#issuecomment-688919240", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { "@type": "Organization", "name": "GitHub", "url": "https://github.com" } } ]

[timmyfaraday]

Random big M generally results in numerical issues... Any quantification on the percentage of cases where it doesn't converge?

[MartaVanin]

not yet, it varies with different pf formulations, too. I made the big M settable in the settings dictionary, so it's easier to play with