Tuesday, July 13, 2010

detection thresholds

Getting to the exciting part of this data structure and histograms: trying to redefine detection thresholds!

Earlier this morning, I plotted my max(S) values onto the image from The Invisibles paper (see below), to 1) verify that I'm getting values that are accurate, to compare values, notice interesting trends, etc.

Walsh, et al (2009)

For the values closest to the ones cited in the paper (nscl =1, pixel size = 1, smoothing length = 4.0), for 1,000 stars and 10,000 stars, I got 6.08 and 5.62 respectively. For all my different nscales, keeping the rest of my parameters the same, I get values that range from 5.25-6.08. In the Invisibles paper, max(S) values lie within that range, so considering that we only ran 100 fields, it's looking pretty good. Changing nscale shifted the numbers more than expected, so that's something to look back over.

Also started changing my program. This time through, I expanded my field size to a 3 degree x 3 degree field, got rid of smoothing length = 6.0, nscale = 1.0, and added pixel size =.25. This new pixel size should slow down computing time, so if it's anything more than a minute for each trial, then I think we'll end up nixing it. I tried the 3x3 field, but it's taking up to much memory right now, so I'm going with a 2x2 field for the time being. If it doesn't work tomorrow, Beth will help me possibly streamline my code so it uses less memory.

My time histogram overplots were really wonky. Note to self: always check images before coming to group meeting. Realized I had just quickly c&p-ed my titles and ranges, and editted the incorrect version. But I got them fixed and ready to go.

There's also a weird trend in some of my histograms. For nscale = 4.0, smoothing length = 6.0, the max(S) is really low. Hopefully this won't continue with a larger field size. If it does, it will be something to closely look into.

Things to do:

- Make cumulative histograms for more in-depth comparisons (can't use plothist --- need to use histogram)
- Run field simulation with 3x3 field size
- See how long trials take with 0.25 pixel size
- Update my latex document (it's been a while...)

All in all, really glad to see that my calculations are looking similar to the ones in the Invisibles paper!

1 comment:

  1. Just thought I'd mention: There is a way to make cumulative histograms with plothist (I'd been doing that this summer), it's just a bit more round-about.