PSF hangs very often
thatsaskguy wrote on Jul 29, 2009
Hello,
I'm using the OGLE dataset and I'm finding that the psf task is hanging very often. It happens on several images, across many different fields. I have not been able to identify anything in common with the fields themselves, or in the way that they are handled by the script I'm using.
This is the output on the screen:
Warning: Star 23287 contains bad pixels outside fitrad
Star 23287 has been added to the PSF star list
X: 1630.85 Y: 1629.47 Mag: 11.811 Dmin: 1940.459 Dmax: 21955.64
21 PSF stars read from 060701_smc0042-7202_3a002_c.pst.2
Trying function gauss norm scatter = 0.07079089
Trying function lorentz norm scatter = 0.06352125
Trying function moffat15 norm scatter = 0.06211673
Best fitting function is moffat15
Analytic PSF fit
Function: moffat15 X: 1022. Y: 1024. Height: 93140.53 Psfmag: 11.305
Par1: 1.048813 Par2: 0.9799262 XYterm: 0.1761353 Moffat: 1.5
<--- and then it just stops running. No errors, no response. It continues to chew processor power (up to 99% of the processor time) but makes no further files.
If you could point me to a solution, or as similar problem that you've solved, I'd appreciate it.
I'm using the OGLE dataset and I'm finding that the psf task is hanging very often. It happens on several images, across many different fields. I have not been able to identify anything in common with the fields themselves, or in the way that they are handled by the script I'm using.
This is the output on the screen:
Warning: Star 23287 contains bad pixels outside fitrad
Star 23287 has been added to the PSF star list
X: 1630.85 Y: 1629.47 Mag: 11.811 Dmin: 1940.459 Dmax: 21955.64
21 PSF stars read from 060701_smc0042-7202_3a002_c.pst.2
Trying function gauss norm scatter = 0.07079089
Trying function lorentz norm scatter = 0.06352125
Trying function moffat15 norm scatter = 0.06211673
Best fitting function is moffat15
Analytic PSF fit
Function: moffat15 X: 1022. Y: 1024. Height: 93140.53 Psfmag: 11.305
Par1: 1.048813 Par2: 0.9799262 XYterm: 0.1761353 Moffat: 1.5
<--- and then it just stops running. No errors, no response. It continues to chew processor power (up to 99% of the processor time) but makes no further files.
If you could point me to a solution, or as similar problem that you've solved, I'd appreciate it.
Mike Fitzpatrick wrote on Jul 29, 2009
This isn't a known problem and without the data and parameters to reproduce the problem I can't really suggest anything concrete. The PSF height of 93K seems a little high to me, is this consistent with the data at all? Does changing any of the parameters (or unlearning the task) help at all, or at least change the behavior?
Because of the spiked CPU usage and the last bit of output you posted I tend to think it's a matter of a fit not converging. We would need to reproduce it to figure out what's happening, but you should start by resetting parameters and incrementally changing them until it reliably breaks, but also check the system log for e.g. a filled /tmp directory or a hung NFS connection if you're using remote data.
-Mike
[/url]
Because of the spiked CPU usage and the last bit of output you posted I tend to think it's a matter of a fit not converging. We would need to reproduce it to figure out what's happening, but you should start by resetting parameters and incrementally changing them until it reliably breaks, but also check the system log for e.g. a filled /tmp directory or a hung NFS connection if you're using remote data.
-Mike
[/url]
thatsaskguy wrote on Jul 29, 2009
Wouldn't you know it, it was something very simple. I was not resetting the datamin value in the script for each individual image. When datamin was above the value of the minimum of the psf function the look-up table would not write. (Possibly it wouldn't converge?) When I corrected this, psf stopped hanging. (I think I wrote that correctly.)
I checked the well depth of the CCD that OGLE used. Its over 100 000 electrons. So, even a high psf would be acceptable and realistic in the data for a bright star.
I checked the well depth of the CCD that OGLE used. Its over 100 000 electrons. So, even a high psf would be acceptable and realistic in the data for a bright star.
Last post on Jul 29, 2009