[daip] autoboxing report

[Eric Greisen]

Lynn D. Matthews wrote:
> The series IM2PARM(4) tests I did were with IM2PARM(1)=1, as I thought 
> it would have an effect on the total # of boxes drawn for the channel. 
> IM2PARM(1)=1 has seemed very robust against boxing sidelobes, so I'd 
> been sticking with that for my first round of experiments. A small 
> change in IM2PARM(3) has now significantly improved the problem channels 
> from yesterday.

That explains the lack of sensitivity to IM2PARM(4).  By taking only 1 
box at most, you probably take one moderately close to the current peak 
value.  Setting IM2PARM(4)=.899 might have shown a slight difference by 
taking no boxes for a cycle more than would have occurred with a lower 
value.  But the net difference should be very small.  You might speed up 
things allowing more boxes at a time so long as IM2PARM(4) is large 
enough to defend against the sidelobes.  My case was full synthesis at a 
good declination in C array L band.

Thanks for your continued help.  Frazer tells me that his student is now 
using FILIT as his main image examination method - but that is 
multi-facet rather than multi-channel so perhaps not of use to you.

Eric

> Lynn
> 
> On Fri, 24 Jul 2009, Eric Greisen wrote:
> 
>> Lynn D. Matthews wrote:
>>>     Eric,
>>>
>>> I finally managed to make a full autoboxed cube of a particularly 
>>> problematic epoch of VLBA data that contains lots of strong, complex 
>>> maser spots. Consistent with my previous tests, the autoboxing worked 
>>> remarkably well on these data, and the results for most channels look 
>>> excellent. This is truly a great advance!
>>>
>>> There were just a couple of small remaining issues. First, regarding 
>>> IM2PARM(4), I attempted to tweak this parameter to see if I could 
>>> improve the cleaning of a group of particularly nasty channels, but I 
>>> find that the boxes and the resulting images are identical, 
>>> independent of its value (I tried values only within the acceptable 
>>> range indicated in the HELP file). This didn't seem right, unless I 
>>> misunderstand how it interacts with the other IM2PARMs.
>>>
>>> Lastly, in the history file, I noticed that a format error occurs for 
>>> NBOX and IMAGR WIN() when there are more than 99 boxes.
>>
>> I attach below 3 HI file outputs - the facet number and box number 
>> were both abused and I have put some effort to make them nice.  The 3 
>> things were done with different im2p(4).  I use im2p(1) = 10 which 
>> will affect what im2p(4) does rather a lot.  With a very low IM2P(4), 
>> I saw the auto-boxing take 6 boxes that were clearly sidelobes of the 
>> box 1 very strong point source.  It also took other things oddly and 
>> excessively at least at the beginning.  With a very high IM2P(4) I 
>> might as well have had IM2P(1) = 1 or at most 2, and it took too few 
>> sources at each cycle.  This would have forced more cycles if I had 
>> set NITER very high and a FLUX to stop things.  I rather liked (but 
>> then I am used to) what it did with the default 0.1 - although a 
>> higher value would be good if the sidelobes are high.
>>
>> Cheers,
>>
>> Eric
>>