Dogs and work (again)

[lederhosen]

Dogs: still adjusting. Dog-Or is nervous of Basil (who tends to growl at him at feeding time) and prefers to lurk upstairs. Unfortunately this extends to not going outside when he needs to do his business, with messy consequences, so we are blocking off the stairs and forcing him to stay downstairs most of the time.

Meanwhile, Basil wants to be friends (at least if I'm reading the body language right) and keeps wandering up to Dog-Or, sniffing him/licking his face, and generally trying to hang around him. Dog-Or gets uncomfortable about this and walks off, so the cycle continues...

Work: busy busy. Being outposted to a different division from Monday, so I'm working on wrapping up my other projects before then. I've been supervising a work-experience student who finishes on Friday. We gave him a small data analysis project (investigating staff attrition) and he's been really enthusiastic about it - apparently he was expecting to be Office Minion getting people coffee etc. My first reaction to this was "people do that to work-experience students?" but then my second was "oh, yeah, that's pretty much what my work-experience was like", so I'm glad we managed to do a bit better. He got to discover the joy that is working with real data (including lots of unstated details like "oh, this data set is censored in a way that's closely related to the variable you're trying to predict") and apparently enjoyed it.

And I'm now the same age that Byron was when he died. Have not become a best-selling poet, but have not been engulfed by scandal either. All in all, would not trade.

Comments

[vito_excalibur]

He got to discover the joy that is working with real data (including lots of unstated details like "oh, this data set is censored in a way that's closely related to the variable you're trying to predict") and apparently enjoyed it.


Actually, that sounds really interesting.

[lederhosen]

Yeah, for all that I grizzle about the quality of the data I have to work with, finding ways to fix that is actually one of the enjoyable parts of the job :-)

[lederhosen]

The issue is: he's working with employee records, trying to predict how long people will stay. When we plotted the survival curve, we found the shape of the curve changed abruptly for people who'd been there more than five years (suddenly the attrition rate dropped markedly).

It turned out to be that the records we were looking at were defined by "everybody who was working with us from mid-2006 onwards". So people who started in the 80s-90s and who are still working with us are on the dataset, but people who left earlier aren't - hence the set is slanted towards the people who DID stay.

Once we figured out what the problem was and fixed it, we got an almost perfect Weibull curve, which is exactly the sort of curve you're supposed to see for survival analysis. I was surprised at how nice it ended up :-)

[freyakitten]

Awesome! :D

[chaos_crafter]

Re Byron - Also, still not dead.