[daip] RMSD bug found
Glenn Morrison
morrison at cfht.hawaii.edu
Tue Jul 22 17:18:03 EDT 2008
Eric,
Another mac-tel compiler (which is free) is at
http://hpc.sourceforge.net/
And they have a special one for Leopard. Have you ever tried it??
--Glenn
On Jul 22, 2008, at 9:55 AM, Frazer Owen wrote:
> Glenn Morrison wrote:
>> Frazer,
>>
>> I'll load the images on my website and send you the link in a bit.
>>
>> The problems with the Mac-intel (does not appear, at least so far,
>> to be a problem with Mac-PPC's) in the past were related to the
>> intel compiler and certain options that were turned on. Since my
>> fluxes and other aspects of the reduction seems to agree with past
>> measurement on linux and Mac-PPC, I think the maps themselves are OK.
>>
>> -Thanks
>>
>> --G
>>
> The problem may go away with "No-opt" for the Intel compiler.
> However, the program takes 4X as long to run on Eric's
> machine. The Linux Intel compiler does seem to be a little different
> from the Mac Intel compiler, i.e. older.
>>
>>
>>
>> On Jul 22, 2008, at 8:58 AM, Frazer Owen wrote:
>>
>>> Glenn,
>>>
>>> It turns out the problem is related to Mac's. Eric gets the same
>>> problem when he runs on his Mac but
>>> not anywhere else.
>>>
>>> I don't believe the 'ROBS' option gets the right answer. I
>>> only trust 'HIST' .
>>>
>>> If you will put the total intensity image somewhere I can get it,
>>> I will run RMSD on my machine and
>>> send you the result. Otherwise it will take some time to fix the
>>> Mac version since Eric has no idea
>>> why the Mac should produce a different answer. This produces
>>> worries about all the AIPS tasks on
>>> Mac's.
>>>
>>> ---Frazer
>>>
>>> Frazer Owen wrote:
>>>> Glenn,
>>>>
>>>> My images simply do not have the problem you report. Maybe your
>>>> problem is related to Mac's.
>>>>
>>>> ---Frazer
>>>>
>>>>
>>>>
>>>> Glenn Morrison wrote:
>>>>> Yo Eric,
>>>>>
>>>>>
>>>>>
>>>>> When I use OPTYPE ='HIST' in RMSD and have the input file as
>>>>> seen below I get the following "quantized" flux plot from IMEAN:
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> However, then I have OPTYPE ='ROBS' the IMEAN plot looks fine as
>>>>> seen below:
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> I ran the midnight job today before I ran RMSD. Both results
>>>>> we're run on an Mac pro mac-tel (AIPS version 31DEC08). You can
>>>>> pull over the radio map from:
>>>>>
>>>>>
>>>>> http://www.cfht.hawaii.edu/~morrison/x1GOODSN/Eric/
>>>>>
>>>>> I've been running RMSD on this map. IMSIZE=100,-1 for both cases.
>>>>>
>>>>>
>>>>>
>>>>> -Thanks
>>>>>
>>>>> --Glenn
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> AIPS 3: RMSD Creates the image of RMS at each velocity plane.
>>>>> AIPS 3: Adverbs Values Comments
>>>>> AIPS 3:
>>>>> ----------------------------------------------------------------
>>>>> AIPS 3: INNAME 'GOODSNV1001' Input image name (name)
>>>>> AIPS 3: INCLASS 'FLATN' Input image name
>>>>> (class)
>>>>> AIPS 3: INSEQ 1 Input image name
>>>>> (seq. #)
>>>>> AIPS 3: INDISK 1 Input image disk unit #
>>>>> AIPS 3: OUTNAME ' ' Output image name
>>>>> (name)
>>>>> AIPS 3: OUTCLASS 'RMS1.7' Output image name
>>>>> (class)
>>>>> AIPS 3: OUTSEQ 0 Output image name
>>>>> (seq. #)
>>>>> AIPS 3: OUTDISK 1 Output image disk
>>>>> unit #.
>>>>> AIPS 3: BLC *all 0 Bottom left corner of
>>>>> input
>>>>> AIPS 3: TRC *all 0 Top right corner of
>>>>> input
>>>>> AIPS 3: IMSIZE 100 -1 Size of the box used to
>>>>> AIPS 3: calculate rms, in
>>>>> pixels
>>>>> AIPS 3: IMSIZE(2)=-1 => the
>>>>> circle
>>>>> AIPS 3: of
>>>>> diameter=IMSIZE(1)is used
>>>>> AIPS 3: OPCODE 'RMS' 'RMS ', 'MEAN', or
>>>>> 'BLAN'
>>>>> AIPS 3: (see help), ' ' =>
>>>>> 'RMS '
>>>>> AIPS 3: OPTYPE 'ROBS' 'ROBS', 'HIST'
>>>>> <<<<<< I ONLY change this!!!!
>>>>> AIPS 3: (see help), ' ' =>
>>>>> 'ROBS'
>>>>> AIPS 3: SCALR1 1 Data<SCALR1*RMS are
>>>>> replaced
>>>>> AIPS 3: by blank
>>>>> (OPCODE='BLAN')
>>>>> AIPS 3: SCALR2 0 >0 =>
>>>>> AIPS 3: ABS(DATA) < SCALR2*RMS
>>>>> AIPS 3: are replaced by blank
>>>>> AIPS 3: (OPCODE='BLAN')
>>>>> AIPS 3: <0 =>
>>>>> AIPS 3: ABS(DATA) >
>>>>> ABS(SCALR2)*RMS
>>>>> AIPS 3: are replaced by blank
>>>>> AIPS 2: (OPCODE='BLAN')
>>>>> AIPS 2: 0 => no blanking
>>>>> AIPS 2: SCALR3 0 Number of RMS used at
>>>>> each
>>>>> AIPS 2: iteration calculating
>>>>> final
>>>>> AIPS 2: RMS. 0 => 3
>>>>> AIPS 2: FLUX 0 Data with RMS>FLUX
>>>>> AIPS 2: are replaced by blank
>>>>> AIPS 2: (OPCODE='BLAN')
>>>>> AIPS 2: 0 => no blanking
>>>>> AIPS 2: XINC 1 Increment at X to
>>>>> calculate
>>>>> AIPS 2: RMS, in pixels
>>>>> AIPS 2: YINC 1 Increment at Y to
>>>>> calculate
>>>>> AIPS 2: RMS, in pixels
>>>>> AIPS 2: The intermediate
>>>>> points are
>>>>> AIPS 2: calculated by
>>>>> interpolation
>>>>>
>>>>>
>>>>> On Jul 21, 2008, at 2:53 PM, Frazer Owen wrote:
>>>>>
>>>>>>
>>>>>> There is another issue which may be important which Eric
>>>>>> pointed out if you use the 16 bit option
>>>>>> for FITTP. The FITS image is scaled to fit between the maximum
>>>>>> and minimum values. If the maximum is
>>>>>> very large while most of the data are very small then the small
>>>>>> values will be quantized. It seems possible that
>>>>>> RMSD could produce a very large number near the edge of the
>>>>>> image. If so that might produce the
>>>>>> quantized result. You should check the minimum and maximum
>>>>>> after you run RMSD to make sure
>>>>>> there are no very large numbers and that you are not using the
>>>>>> 16bit option for FITTP.
>>>>>>
>>>>>> ---Frazer
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> Glenn Morrison wrote:
>>>>>>> The RMSD map I sent you was not made using HIST.
>>>>>>>
>>>>>>> also Frazer suggests IMSIZE=100,-1 to get a circular region. I
>>>>>>> would not use any option but 'HIST'.
>>>>>>>
>>>>>>> Not sure what problems Edo has had. Is he using the latest
>>>>>>> version of AIPS?
>>>>>>>
>>>>>>> -Thanks
>>>>>>>
>>>>>>> --Glenn
>>>>>>>
>>>>>>>
>>>>>>> On Jul 21, 2008, at 1:02 PM, Rob Ivison wrote:
>>>>>>>
>>>>>>>>> --Big thanks to Edo (what is his email address??) for
>>>>>>>>> finding this!!!!
>>>>>>>>
>>>>>>>> Edo is ibar at roe.ac.uk and he's a real gem. not sure i've ever
>>>>>>>> found anyone quite so thorough. he makes Biggs look sloppy,
>>>>>>>> and when i first met Biggs i thought he was beyond retentive.
>>>>>>>>
>>>>>>>> Edo informed me today that a noise map made using the HIST
>>>>>>>> option in RMSD plays havoc with SAD, amongst lots of other
>>>>>>>> little snippets of wisdom. sometimes it feels like he's
>>>>>>>> supervising me.
>>>>>>>>
>>>>>>>> cheers,
>>>>>>>>
>>>>>>>> Rob
>>>>>
More information about the Daip
mailing list