Problem solved everyone. Britton and I tracked it down. When you make a Homogenized Volume, you need to set the field explicitly that you are homogenizing. By default it uses the "Density" field. So effectively, I was comparing a projection of "Density" (from the homogenized volume method), and a projection of "CIV Number Density" (from the KDtree method)--of course these would differ. This just goes to show you that one should always RTFM prior to emailing the list with problems. Sorry everyone! Cameron On 11/22/11 3:31 PM, Cameron Hummels wrote:
Hey Sam,
When I run each method over the whole volume, the kd-tree and the HV take the same duration to process, however, the ratio of the two is still the same factor of 2e-8.
Cameron
On 11/22/11 3:26 PM, Sam Skillman wrote:
Hey Cameron,
Are the answers similar if you do the entire volume? The kd-tree can not accept things like spheres to homogenize over, so maybe it is because it is projecting the entire box? I'll keep thinking...
Sam
On Tue, Nov 22, 2011 at 1:07 PM, Cameron Hummels <chummels@gmail.com <mailto:chummels@gmail.com>> wrote:
Hello peeps (mostly Britton, Matt and Sam),
I have recently been doing some off-axis projections in my cosmological runs (using the supercool new off_axis_projection helper function Matt wrote), and I've encountered some problems. I find different results when I do the off-axis projection using a homogenized volume versus when I do not use a homogenized volume (when it uses the default behavior for camera objects -- ie a KDtree).
Of course, these two results should be identical, and they are when I use a normal field like "Density". However, I'm trying to use a derived field from some code Britton wrote, part of a package called ion_balance, which creates derived fields for different atomic ions. So when I compare the CIV Number Density from these two methods, I get very different results. Even when I do this on a normal vanilla yt field, like "Density", the KDtree method takes exceptionally longer than the homogenized volume method (I think this is because I'm only doing the HV for a small subsample of the overall volume). On the other hand, they both take about the same amount of time when my sample volume is the entire box volume.
I've pastebinned a demonstration script which shows this discrepancy at: http://paste.yt-project.org/show/1953. If you don't have ion_balance, you can comment that import out, and comment the line for defining the field as "CIV_Cloudy_eq_NumberDensity", and run it to see the time discrepancy between the two methods. It should work on any sort of parameter file, not just the specific one I'm using. What I do is take an off-axis projection using each method, then divide the two images against each other to form a ratio image, and then output the average and stddev for this ratio. The average of the ratio is: 2e-8.
I've changed the width of the off-axis projection and it has a minimal (but nonzero) change on the overall ratio between the two.
So I'm not sure what to do. It appears that the CIV field is initiated in the same way that a normal field is, with the projection_conversion set to 'cm', just as it is for "Density". Any ideas on what could be making this difference? Any ideas on which is the value to trust?
Cameron _______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org <mailto:yt-dev@lists.spacepope.org> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
_______________________________________________ yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
yt-dev mailing list yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org