[ml] Statistics / software question
gershon.bialer at gmail.com
Sat May 12 01:06:13 PDT 2012
I think you can use Kalman smoothing (see
This should give you an estimate of the value and standard deviation
for each point, which I think you should be able to use for confidence
R has a function KalmanSmooth, which might work. You could try a
simple model with T=Z=1.
On Fri, May 11, 2012 at 10:37 PM, Christoph Maier
<cm.hardware.software.elsewhere at gmail.com> wrote:
> The NaI(Tl) or better yet, CsI(Tl) is the stuff hard to get by.
> We need to find a Jeriellsworthesque way to cook that in our own kitchen, or
> at tastebridge.
> Looks like it's time for my (nowadays) semi-annual dose of noisebridge
> reality distortion soon.
> Data: CSV, if you please.
> On Fri, May 11, 2012 at 10:30 PM, Jake <jake at spaz.org> wrote:
>> well i've stashed graphical plots of the bins, but I do not have stored
>> raw data at this time because my java program what makes these graphics
>> doesn't save the serial data to a file. oops.
>> But i will correct that asap and start providing data. in the meantime
>> you can look at these plots:
>> and you will notice that there's absolutely no data there, it's all noise.
>> I have determined that this is because my scintillating crystal of NaI(Tl)
>> is "fried" meaning the person who sold it to me was an asshole, and the
>> amount of blue light coming out of the crystal has nothing to do with the
>> energy of the original gamma. Random data.
>> fortunately I have other crystals (including a BGO crystal) and I will
>> capture more data soon, which will hopefully look more like this:
>> there is other data, from an older version of this detector with a
>> different crystal (a nasty ruined crystal but maybe not quite fried)
>> which you can look at here:
>> On Fri, 11 May 2012, Christoph Maier wrote:
>>> Link to the raw data, please.
>>> On Fri, May 11, 2012 at 8:55 PM, Wladyslaw Zbikowski
>>> <embeddedlinuxguy at gmail.com> wrote:
>>> Hi, my friend Jake has a little project for which we might be
>>> able to
>>> use some statistics expertise.
>>> We are taking analog readings from a device (a voltage pulse).
>>> voltage pulse represents the energy of a single X-ray particle
>>> MeV). We know what the energy signature is supposed to look like
>>> this particular radioactive material; i.e. there are peaks where
>>> certain energies are highly represented, and valleys where other
>>> energy levels are rare. So we would like to correlate our
>>> with the expected signature.
>>> The problems are:
>>> 1. A lot of noise. We have a signal:noise ratio around 1:1 or as
>>> as 4:1, because of background radiation and attempts at
>>> 2. We don't know exactly how the voltage we read maps to MeV.
>>> Voltage is a function of Energy, presumably linear, but we don't
>>> exactly the scale (how many MeV per volt).
>>> SO in short, we have a graph of our data, and we want to
>>> force-fit it
>>> to the graph we expect. My idea is to apply noise removal and
>>> getting the closest possible match. Any thoughts on this? R?
>>> Possible topic for a meetup? We can post the graphs and the
>>> if anyone is interested to see.
>>> Thanks in advance!
>>> ml mailing list
>>> ml at lists.noisebridge.net
> ml mailing list
> ml at lists.noisebridge.net
More information about the ml