@Raphael of course, but afaik food delivery is non-international, so donuts either here (or very special local cheese-cream pies) or provided address of your local sweet-shop that accepts remote payments ;)
@Raphael so you have seen my awesome question, didn't answered, didn't upvoted and didn't even closed? How not-clickable of you :-/
Even edit or comment...
I got question about attribution on paper - I did some review, cheering up, input, pointed some references - nothing big. I denied attribution and pointed that "too big group of participants" will make the author credit seems smaller. I got response that more is the merrier, because the paper was seen and reviewed by more people so it's value increases. I do not fully agree, but how does it look really?
@EvilJS I never thought about attribution in that way. Those who co-write the paper are authors, and nobody else. Those who contributed in other capacities get a mention in the acknowledgements section, and nobody else.
It's not an issue of what looks good, it's an issue of adacemic integrity.
If you did only the final write-up and all ideas came from elsewhere, sure that doesn't "look good" in the sense that people will praise you a lot. Still, you gotta give credit where due, and not take more than you have earned.
Note bene: I'm fully aware that the rules are routinely violated, in particular by senior researchers who insist on being co-author on every paper their group/lab produces. That's highly unethical, imho, and should neither be copied nor excused or supported.
(There are lots of threads on Academia on these issues.)
It sems that imgur is recoding JPEGs, but does not touch PNG. So some LZMA packed file stored as raster and compressed to PNG to make permanent external data storage seems fine.
The question Proof of correctness for the greedy solution of uva10249 The Grand Dinner is presently on hold. It is definitely not self contained, so I understand it needs editing.
My question concerns the remark by @Raphael: "It's inappropriate to outsource contest questions". Let me start by sa...
@EvilJS nice find/ paper. nearly an answer to that PCA question dont you think? yes sec VII nice summary of general applicability. have you tried/ used PCA? it can be seen as just "normalizing/ centering/ scaling" multidimensional inputs.
@vzn of course I used PCA, all types that I know exist. I just treat all statistics with caution.
@Raphael there are very hard optimizers that trear "lossless" with some relaxed way to make it smaller. So idea to compresd more to save space is not that bad. But recoding JPEG to be about similar size enhances artifacts.
@vzn this is one of contrarguments of using PCA to find k in k-means. If it fails (and often does) it is useless, but when it works, then it is weaker version of k-means, so calculating it for k and then discarding resultsis a waste, when it works k-means with high probability returns the same results, so then (and only then) we can just feed results to actual centroids and we are almost done.
Also the initial step of some ICA techniques is to subtract PCA (when the noise is guaranteed gaussian it makes perfect sense). I found such cases and people who can use it that way lucky. I do not have that luxury.
How many arbitrary large samples are needed to destroy mean? 1. How many to destroy median? 50%. Cool. My daily data is more than 70% noise and although hostile environment, this is not even the main issue.
