Stats weirdness [fixed :)]

[Scarblac]

The overall project stats graph is going upwards nicely, but the last 24-hour graph *always* slopes downward.

I assume that it's somehow adding in results that came late to the graph, but this makes the 24 hour graph rather misleading.

Can someone explain?

[MDFaunce]

This is my understanding, so take it as offered:

When you turn in a result, your computer says something like:

I tested (k,n) and I started on mm/dd/yy hh:mm:ss and just finished it.

The stats server then spreads your work over the entire time period.

The longer it takes to perform a test, the worse that last 24 hour graph is going to look. It's delayed 6 hours already, and I'm not so sure that even delaying it 24 hours will really help.

What would make more sense (to me) is a last week graph with the most recent 24 hours removed (ie, 8 days ago to 1 day ago)

[smh]

Yes, in the last part of the graph you'll miss all the calculations done since the last connection to the server for each computer running the client. This adds up to quite a fe cEMs.

[Stricker]

it also depends on how slow your computers are i have 2 computers that turn their data in almost every 30 minutes but 1 computer that turns its data in evey 5 hours and 2 other computers somewhere in between

so in my case the stats graph only "misses" data over the last 5 hours or less not over a major segment also since that slow machine only makes up about 13k of my data it is barely even noticable on my graph compared to 3 machine reporting between 95-140k each

[jjjjL]

I changed the last day overall rate graph. It is now one week (with a day delay). I basically took MDFaunce's suggestion since it was a particularly good one.

It looks much better.

The reason the last day graph is constantly downsloping is actually incredibly complex. It's a little deeper than what MDFaunce suggests (although he is correct about the spreading). Basically, I believe it is a combination of the spreading and the fact that it is more likely that longer tests will have a period where they are run on computers that are turned off.

I didn't really like the appearence of the downward slow phenomenom any more than anyone else because it was incredibly deceptive and made it look like the project was continually contracting. I don't think SB has undergone contraction at any point in the past 3 months, much less in the last day. Considering we had 30 new signups today and have grown to over 2000 signups, we don't show any signs of slowing down in the near term.

Check out the new graph and let me know what you think:
http://www.seventeenorbust.com/stats/

-Louie

[MDFaunce]

I sure hope this isn't my "15 minutes of fame."

But, I like it

Of course, it was my idea !!!!

Back to lurking.

[Scarblac]

Originally posted by jjjjL
[B]I changed the last day overall rate graph. It is now one week (with a day delay). I basically took MDFaunce's suggestion since it was a particularly good one.

It looks much better.

Indeed, good work :-)

[Nuri]

New graph is much better.

May be we should also think about a one day delay in the last 3 months graph as well. as far as I understand, the same rule applies there too.