Hi Nate, I've never used AverageDeviation. If I recall, I read about it on the twiki and it's kind of random, could be useful but not really a standard stat, so if it's super broken then I think it's ok to nuke it.
I like the Distribution stats, I've been using them a fair amount in my tree, which is why I fixed up VectorDistribution to work correctly. But the bin thing you describe great - one thing that is annoying about Distributions is that you have to run it in order to figure out how to set your bins, especially if you have no intuition about how things will be distributed. Would it be possible to be able to have both? I know with matplotlib, when I plot a histogram it's often sufficient to let it auto-generate my bins, and even if i want more granularity, i just make more bins...but sometimes I want to compare two things, and the auto-generated bins between the two things are different. In this case, I appreciate the ability to manually set bins to do a bin-to-bin comparison. That would be useful in this case as well...particularly if these values were to be pulled out to be plotted between runs, it would certainly make for an easier comparison if the bins are the same. But then, that again requires prior knowledge of rough bin distribution, which is what your auto-binning is trying to get away from. So...I'd say both models are useful. It would be great if both could be supported, just depending on how the stat is inited. Lisa On Thu, Jun 24, 2010 at 6:43 PM, nathan binkert <[email protected]> wrote: > This message is mostly to Lisa since she's done some stats cleanup > recently, but anyone who has used statistics other than Scalar and > Vector should feel free to comment. > > I'm doing further cleanup and I'd like to know, which stats do you > use? Do you use AverageDeviation (it's broken)? Should I fix it and > keep it, or can I just delete it? What about the various other > Distribution type stats? I've created a new Histogram stat that's > like Distribution. It's needed by Ruby, but is generally pretty > useful and possibly more useful than Distribution. Basically, in this > stat there's a fixed number of buckets, but the bucket size grows to > accommodate all of the samples. If there's an outlier, unfortunately, > it could pack everything down into a small number of the buckets, but > other than that, it relieves you of the burden of figuring out how big > buckets should be. > > Anyway, I'm going through a cleanup of stuff and I want to know what I > need to keep and what I can drop. > > Nate >
_______________________________________________ m5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/m5-dev
