Is the uniform histogram type wrong? about folsom HOT 7 OPEN

Oh, looks like this is how Vitter's Algorithm R works. So it is more a question of specification than one of implementation mistake.

from folsom.

Thanks for checking @jlouis. @russelldb might have some thoughts on how this would effect slide_uniform.

from folsom.

I need to look into it too. It seems @Vagabond made this change 2be6249#diff-b7a6cde361f08ae87401c3a98eb116c7 that changes the behaviour of the slide_uniform sample.

from folsom.

Ah, it looks like @Vagabond actually fixes slide_uniform as it was using Size not MCnt as the upper bound for the random number generation (i.e. a fixed, rather than growing value.)

So is the plan to change the sample algorithm?

from folsom.

The code as it works now does the right thing. A uniform histogram samples its reservoir over the entire set of inputs. So if you have a million inputs already, the chance of replacing into the sample reservoir is 1/1000. Which is what you want for the complete overview.

Most of the time however, you want some kind of histogram window in which case the slide_uniform or exdec solutions are more appropriate. So it is really a questions of "what do you want". Not a question of what is correct or wrong.

The reason I opened this was because I was getting unexpected data. But it turns out the data are correct and my expectations of what uniform does was wrong :) You can close this one.

from folsom.

@russelldb I don't think we should change anything. But we should definitely consider explaining what the histogram types does in the README :)

from folsom.

Folsom has moved, please resubmit your issue at https://github.com/folsom-project Thanks!

from folsom.