Turns out that the changes described in my last post didn't completely work.

First, the use of time entropy in the drifting RFI algorithm (see last post) was the right idea, but there were cases that actually were RFI but for which the entropy was very low, so we were failing to remove it. Dan W. pointed out some of these. I switched to a different approach where we divide each triangle into 10 time intervals; if signals are present in < 4 of these intervals, we don't count it as drifting. This fixes the cases that Dan found; hopefully it works in general.

Second, there's the recurring issue of how to reward (i.e. boost the score of) multiplets whose signals are spread out across the whole range of times we observed that sky position. Or conversely how to penalize multiplets that consist of two signals a second apart. This is key to SETI@home - we're the only radio SETI project that looks for signals that recur over years and decades, and we need to maximize this advantage.

The multiplet score includes a term called "nd_factor" (ND stands for "number of detections"; detections is what we used to call signals). Without going into mathematical detail:

• Our formula for nd_factor was outdated and wrong.
  
• Eric proposed a new formula, which I implemented (perhaps incorrectly) and it had the opposite of the desired effect. We knew this because birdie scores got worse - thank goodness for birdies!
  
• I proposed a somewhat simpler formula (i.e. simple enough that I could understand it), and it seems to work well. Also - thanks to an enhancement by Eric - it also subsumes the "frequency factor" (rewarding multiplets that are close in frequency) so we don't need a separate term for that. And finally, because the formula rewards multiplets with lots of signals, it seems to do what we want in terms of suppressing 2-signal multiplets.
  

This is not necessarily the end of the story, but I think we made a step forward.

One other thing: after each run I create a "snapshot" consisting of a few key files, such as the list of multiplets. The idea is to give us a way of comparing the results of different versions of the algorithms. However, there wasn't a convenient way of looking at old snapshots. I added one: when you click on a snapshot you see a page that looks like the main Nebula page, but for that snapshot. Not everything works - e.g. there are no waterfall plots, because we don't save that level of data - but at least you can see the multiplet list.

This is not necessarily the end of the story, but I think we made a step forward.

+1

---

A proud member of the OFA (Old Farts Association).

With these various iterations of Nebula, are you beginning to see consistent regions of the sky that get flagged as interesting for follow-up?

---

a number of tables have been published giving the RA & DEC of areas covered and the signal found in them - one example of these tables is:
https://setiathome.berkeley.edu/nebula/pixels.php?s6=
Using one of the many on-line star maps you should be able to plot these locations and see hot-spot.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Thanks Bob
I've seen these tables before for each new version of Nebula. However, I'd never taken the time to see if there are any regions that are consistently high scoring (which is the whole point of what Nebula does). Guess I was jumping ahead with my curiosity before any paper/announcement is made.
Jon

---