MOEA, HillClimbing and Our Method

[ginfung]

What happened?

• Our Method = My prev code -Manhattan distance + Euclidean distance + cosin projection
• HC = replacing our surrogate method with Hill Climbing

[ginfung]

HC is not good. SA is better than HC. -> our problem is NOT convex. SA can handle un-convex problems. (reference https://www.quora.com/What-is-main-difference-between-hill-climbing-and-simulated-annealing)

[timm]

is this where we are right now? there's a few cases about were green is much worse than the others.

is this with 20+8? or with 30 random?

FYI, id go back to manhatten

[ginfung]

manhatten, sure thing.

now its 20+8. 30 is not good enough.

i am working at real machines. but there are so many configuration(devops) issues to handle. because each workflow is treated as a single software and has its own deploy ways.

Sent from my mobile device Jianfeng Chen Department of Computer Science North Carolina State University Raleigh, North Carolina http://jianfeng.us/ 919-457-2034

On Dec 29, 2017, at 11:02, Tim Menzies notifications@github.com wrote:

is this where we are right now? there's a few cases about were green is much worse than the others.

is this with 20+8? or with 30 random?

FYI, id go back to manhatten

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[ginfung]

Manhattan distance applied in following charts

[timm]

thanks. when can i see the above as the following table? may beed seperate tables for spread, hypervolume, etc

[remark- image above was from icse submission]

[ginfung]

Quality measures

[timm]

so that's all good

all you need now is a case that your predictions are rank preserving.

now you need to do a lot of stuff with patrick in jan to get that nsf project powering on

so i need to ask you... what is the least effort thing you can do to check rank preserve? forget frank's lab and try some AWS instances? make a case based on some prior CLOUDSIM study?