Ten fold - 2010-12-10

« first day (115 days earlier) ← previous day next day → last day (4886 days later) »

1:16 AM

Looks like I missed the chatters again... It's rather misleading, though, as it seems to imply there are 14 people here, when what it means that there are 14 people in all rooms (not even sure if this means other CV rooms, or just all rooms on the entire site).

Back to my previous post, I'm teaching myself DLMs via R's dlm package and finally made some progress. I've got to say that DLMs are WAY do-it-yourself, some-assembly-required, batteries-not-included kind of tools. Or maybe more like making your own paint, paint brushes, and canvas before you start painting.

Or, maybe like making your own lightsabre... though it's also easy to cut your own hand off during the process. I like it a lot and it really has let me analyze my time series in ways that ARIMA tools did not, but in the end you really do have to keep columns of matrices straight and if you reorder something slightly you could be looking at seasonal slopes instead of levels and not see your mistake at first glance.

13 hours later…

2:09 PM

@Wayne I'm not an expert on DLMs, but I did skim through "Dynamic Linear Models with R" and would recommend it if you're getting started.

amazon.com/Dynamic-Linear-Models-Giovanni-Petris/dp/0387772375

2:20 PM

Hey @chl!

Hi @Shane!

@Shane It's been long that I didn't visit the chat room, sorry about that...

2:52 PM

@chl Yes, it would be good if we could start to generate some traffic in here again.

@Shane Yep, I agree. @mbq ask for that on meta, no?

I tend to avoid reading anything that @mbq says.

How about the Polystats project?

3:09 PM

Good point. That just dropped off, didn't it. I periodically think about it, but have just been so busy.

Didn't you setup a wiki or something like that to hold all stuff related to this project? There were also some discussions around data dump (meta.stats.stackexchange.com/questions/463/…). Any progress on that?

Anyway, I think it will be interesting to have some stats on users traffic and Q&A activities, much like @csgillespie did (some 4 or 5 weeks ago).

I agree. And making it a polystats project gives it a little cache.

The data was downloaded, and I created a wiki (statalgo.com/polystatsproject/wiki) which can be freely updated, etc.

But Srikant was really driving it, and it stopped immediately following the data dump.

Do you want to pick up the ball and run with it?

@chl Beware, this idea is cursed.

All Srikants that were doing it suddenly stopped.

@mbq may be right. This site is a hobby for most of us. The polystats project could start to feel like...work.

I have to think about that (I'm not really an expert of SE, I'm just on CV); @whuber also had some ideas I think. Anyway, we can try to revive it in the near future: As all of you mods said, questions tend now to be more related to statistics than to "how do I produce an XY plot"; I think expert users in related fields (psychology, biology, maths, etc.) will be able to contribute and enhance CV quality.

3:23 PM

Yes, I do like the idea of taking the strong community of the site and using it towards some larger goal, but it's definitely hard to orchestrate, and will require some time/attention.

Obviously, if it becomes to resemble work, it won't work...

Right. Maybe we could get some fun prizes (mugs, mousepads) for those who participate. We can ask Robert Cartaino if they can fund that...

Having a little swag can always act as a motivator.

But definitively, things are going better since the beta; there are not just plain readers, and there are some regular users; questions/answers are updated; that all sounds good to me. Still newcomers have to wait to earn enough rep to fully participate.

Good idea.

Good idea

Can we imagine having some "mini-polystats" projects (with well defined questions) that users can work on, alone or in small group, without any time constraint; results could then be summarized and discussed on meta or here, and those users would earn rep+badges for that?

3:33 PM

Of course, anything is possible. I suspect that having badges for it may not go over so well, but it doesn't hurt to ask.

Do you have an idea for such well defined question?

Also we could call the badge Percolator (-;

Ha

e.g., reputation effect (are high rep users favored in some way, especially when there are concurrent and equally good response), tag clouding, ...

I'm sorry, I have to go -- let's talk about that later.

Tags formed a one giant blob cluster in my first approach to that problem.

But the first one is interesting. SE team claimed on meta that they've checked it and the answer is that there is no correlation.

@chl I've actually done another analysis of all SE exchange sites that have come out of beta. I've just not had time to post the results and data. I'll try and do it early next week.

4:18 PM

@csgillespie Well, it seems the mug-giving idea worked pretty nice (-;

@csgillespie Good to know, thanks for all your work!

4:32 PM

If it is so crowdy here, what do you think of the journal club idea?

2

3

Q: CV journal club?

We have an awesome chat (yes, it is awesome by design), and a small traffic there. Yet there is an idea to organize a periodic meetup to boost the chat popularity. Photo.SE has for instance photo critique sessions; we could have a some kind of journal club. So, two questions: Do you think it ...

discussion chat

I like it.

We can schedule it as an event: chat.stackexchange.com/rooms/info/18/ten-fold?tab=schedule

me too

Good idea @mbq

Ok, so maybe some idea for the time of the first?

looks fine to me too

Hmm...weekend or weekday?

4:41 PM

@chl As @Shane mentioned when it becomes work, it drops to the bottom of a very long list. It's still fun just now ;)

@mbq I think weekly is too much. A stats paper can be tough going!

@Shane Friday?

@csgillespie I also think weekly is too often.

@csgillespie Yes, I'm aware of that. We cannot devote too much time or feel like having some constraints, otherwise it will be counterproductive in every aspects I can tkink of

@mbq Monthly I think would be enough

@csgillespie Agree

Ok.

So what about 17th?

4:45 PM

Just added my tuppence on meta. Summary; great idea

@chl @Shane Regarding analysing the SE data, the most efficient method would be to access the data dump. The API isn't designed for downloading vast quantities of data. In my last experiments I had to throttle the connections to one request every 0.3 seconds.

@csgillespie I fetched the data one month ago but in the meantime I changed my computer and forgot about that

@GavinSimpson The access issue is a good point; yet there are a lot of open access papers and many of closed-access have a preprint somewhere on the internet.

@csgillespie There you go being too practical again.

@mbq yes, we need to be as inclusive as possible so sticking to open access or preprints would be a big plus

4:51 PM

If we're going to discuss preprints, we should be slightly careful about notifying the author since this will be indexed.

Not sure about the most appropriate way to deal with that...

@Shane Invite the author?

@csgillespie And maybe give them the ability to "opt-out"?

On the other hand, if it's on the internets, then it's somewhat out there for discussion...

I just think that it serves us to be cognizant of people's reputation, etc. when doing this.

Another option, at least for the first one, would be to discuss a "classic" paper.

@Shane if the author came along, then it's free publicity for their paper. I presume that we won't find any fundamental flaws in the paper. What's more likely is we misunderstand a few things.

One that was groundbreaking in some way.

I guess most authors that are actually publishing in OpenAccess journals will be happy to participate if they have time

4:53 PM

Absolutely.

@chl So it seems it could be a good way of promoting the site.

When you say OpenAccess, you mean something like SSRN or arxiv, right?

I had in mind BMC on biomedcentral

Someplace where people can freely publish preprints without peer review, etc.

but other eprints repos are probably fine too

4:56 PM

@chl JMLR, JStatSoft...

As a separate idea, we could also have an "R Package Demo" session, where an R package author would come in and demo/discuss their package.

2

I'm still puzzled by the fact that Hadley Wickham, Andrew Gelman, Doug Bates, etc. don't come there

@Shane Yes, good idea

@mbq yes, JMLR or Cobra Biostats (biostats.bepress.com/)

@chl Hadley comes here for time to time (last seen 2 days ago).

@mbq yes, I know: from time to time...

Ok, I definitively have to leave, talk to you later!

Have to head. Catch you all later (won't be back until Monday).

5:04 PM

@Shane I think we meant get the preprint for a published paper if the journal/publisher version were not public/open access.

I should have said author version or reprint rather than preprint. sorry.

@Shane Superb suggestion!

@GavinSimpson Good, because there's a lot of crap out there, when there's no bar for entry.

I find that to be especially true on arxiv.

@Shane Indeed...

5:30 PM

Ok, so it seems we would also need a meta question to announce this event and ask for a propositions of articles to discuss; then we'll try to contact the author and ask if she/he would like to participate. For now I'll stick to 17Dec, if it won't be enough time it can be always postponed.

« first day (115 days earlier) ← previous day next day → last day (4886 days later) »

Transcript for

Dec '1010

Ten fold

CrossValidated's general room for gossip, grumbles, and idle c...

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC

site design / logo © 2024 Stack Exchange Inc; legal

mobile