My last post described how we're narrowing the range of non-barycentric multiplets we look for, namely to those consistent with being from habitable planets orbiting F and G type stars. This was very successful; I did a run of the Nebula pipeline, scoring 250K pixels, and we "found" (in the sense discussed below) 357 out of 2000 birdies, whereas before we barely found any. Given that we're generating non-bary birdies with extremely high powers, this isn't that great. But it's a big step forward. We'll see what happens when we reduce the power.

I added some critical info to the web display. We no longer have a single "score" for multiplets; instead, we have 7 "score variants", namely the combinations of the three score factors. Similarly there is no longer a single "rank" for a multiplet; rather, there are 7 ranks, one for each score variant. We consider a birdie to be "found" if, for any of its multiplets, at least one of these ranks is 100 or less.

I added all this info to the web pages. The birdie list now shows the best rank for each birdie; you can see at a glance whether it was "found". At the bottom of the list I show how many were found. In the detail page for each birdie, you see the score and rank variants for all of the birdie's multiplets.

Eric has continued to fix bugs in generating birdie spikes. This has turned out to be a very complex piece of code. But I think he has it pretty much wrapped up.

I've done a couple of things kind of for fun. First, I looked my algorithm for "time overlap pruning" - taking a set of detections, and choosing a subset that doesn't overlap in time, and that maximizes the total "value" of the detections (for spikes, value is the same as power). I was doing it in a simple way: scan the detections in time order; if consecutive ones overlap, keep the one with higher value. This isn't optimal, and in certain cases it can be very non-optimal. I did some web searching, and I discovered that there's a well-known scheduling problem called the "Weighted Activity Selection Problem", and there's an efficient optimal algorithm for it! It uses a technique called "dynamic programming" where you build up a solution from smaller sub-problems. I was very excited about this and wrote a slick C++ implementation.

Secondly, in keeping with my "more visualizations" theme, I wanted to see sky plots: for example, how the detections in a birdie look on the sky. In other words, to plot them by RA and dec rather than time and frequency. So I implemented this. The web pages for pixels and birdies now have "Plot RA/dec" buttons that produce a sky plot. For pixels, this shows the spikes in the disc for that pixel, which includes parts of the 8 adjacent pixels as well. Here's an example:



This pixel contains a birdie, so there are birdie spikes as well as real spikes. The spikes line up in rows; these are the passes that beam centers made through the disc at different times (possibly years apart). Some of the rows are horizontal; that's when the telescope was "drifting" and the beams were moving in RA but not dec.

You can also look at the birdies in multiplets. Here are some examples:







In some of the multiplets, the spikes are in tight rows; in other's they're sprinkled randomly. I'm not sure why; I'll need to discuss this with Eric.