Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.atnf.csiro.au/research/WALLABY/Wallaby2011/SIM-FEST2011-RJurek.pdf Äàòà èçìåíåíèÿ: Thu Dec 8 08:17:50 2011 Äàòà èíäåêñèðîâàíèÿ: Mon Feb 4 09:03:28 2013 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: ï ï ï ï ï ï ï ï ï ï ï ï ï ï

Image Credit: Gyula Jozsa

Leveraging visualisation for WALLABY
Russell Jurek WALLABY Simulations Fest 2011, November 23rd 2011


WALLABY/DINGO WG5
· · · · · · · · Baerbel Koribalski (PI) Lister Staveley-Smith (PI) Martin Meyer (PI) Russell Jurek (co-chair) Chris Fluke (co-chair) Amr Hassan Andreas Wicenec Gerhardt Meurer

· Fortnightly meetings starting soon


Talk Outline
· The trick to exploiting visualisation · Visualisation uses
· · · · · Data mining Datacube quality control Source finding Testing parameterisation Source classification

· Citizen science · Examining weird objects · Summary


The trick


The trick
· Visualisation is good for rapid, qualitative analysis
· Visualisations are contextual and `information dense' · Visualisations provide perspective

· Images are engaging · Leverages human pattern recognition capabilities · Visualisation is the key to citizen science


Visualisation uses


Visualisation uses
1. 2. 3. 4. 5. Data mining Datacube quality control Source finding Testing parameterisation Source classification


Data mining
· Exploit context
· Easiest way to spot the weird and wonderful

· Exploit perspective
· Identify large scale structure · Identify selection effects

· We need to be creative
· Cluster finding?

Which of these is not like the others?


Visualisation uses
1. 2. 3. 4. 5. Data mining Datacube quality control Source finding Testing parameterisation Source classification


Datacube quality control
· 2 levels of datacube quality control · Internal QC
· Part of ASKAP pipeline · Automated tests with quantitative results

· External QC
· Semi-automated and manual tests · Incorporates visualisation · Addresses weaknesses of automated tests


Internal QC
· Measure global noise level of datacube · Test sensitivity/noise variation of datacube
· Measure RMS in running box · Compare to sensitivity model and survey requirements

· Test continuum subtraction of datacube
· Compare continuum subtracted and blank line-of-sights · Are the voxel flux distributions the same?
· Apply Kolmogorov-Smirnov test · Compare medians and interquartile ranges

· Test for presence of known sources · Test if source count and distribution is sensible
· Use predicted counts and predicted variance

· Test if source parameters are sensible


The problem with automated tests
· Measured quantities describe a macroscopic state
· Potentially large microcanonical ensemble · Scale dependent

· Relies on picking the right test · Effects might be intermittent
· Another mechanism for scale dependence


The problem with automated tests


The problem with automated tests


External QC
· · · · · Goal is to address issues with automated QC tests Make use of low resolution datacube Visualise datacubes or datacube sub-regions Visualise running RMS of datacubes Overlay catalogue, known sources and sky model on datacube visualisation · Visualise calibration residuals
1. Measure calibration residual for known sources 2. Plot as a function of position in the datacube 3. Look for trends/distinct regions


Visualisation uses
1. 2. 3. 4. 5. Data mining Datacube quality control Source finding Testing parameterisation Source classification


Source finding
· Citizen science? · Best on low resolution datacubes? · Procedure
1. Remove sources from datacube 2. Examine datacube for oddities/sources


Visualisation uses
1. 2. 3. 4. 5. Data mining Datacube quality control Source finding Testing parameterisation Source classification


Testing parameterisation
· Citizen science? · Show overlap of sources and models/masks:
· · · · · C C C C C o o o o o mp mp mp mp mp a a a a a re re re re re TiRiFic model projection and cubelet velocity field and model velocity field rotation curve and model rotation curve source finding binary mask and cubelet integrated spectrum and cubelet integrated spectrum


Visualisation uses
1. 2. 3. 4. 5. Data mining Datacube quality control Source finding Testing parameterisation Source classification


Source classification
· Citizen science · Repeat galaxy zoo with WALLABY · Additional classifications:
· Multi-wavelength overlaps
· Degree of overlap · Degree of overlap correlation

· Check/Find object merging


Citizen Science


The benefit of citizen science
· 1,500,000 sources = 13 FTEs (using CSIRO definition) · 1 FTE = 8 FHRs (5 hour regulars, 5hr/week for 48 weeks) · 1,500,000 sources = 104 FHRs
Eyeballs 10 10
6

FHRs 70 695

7

14.41 Million 1000 10
8

6945

MANY HANDS MAKE LIGHT WORK


Implementation
· Most of our operations are comparisons
· We can easily set up fuzzy answers · A single tool is possible

· Server that provides HI images and a random overlay · Overlays correspond to different comparisons · Generate catalogue of classifications
· Identifies the weird and wonderful

IT'S FEASIBLE EXCITING


Examining weird objects


WALLABY HI Analysis Tool
· What do we do with the weird stuff?
· HI poor and HI rich galaxies · No obvious multi-wavelength counterpart

· Let's do the simple thing
· Leverage citizen science to refine the list of weird things · Look at every object in this (hopefully) small list

· WALLABY HI Analysis Tool (WHAT)
· · · · · Single source analysis tool Single source visualisation Multi-wavelength overlays S'Finder mask overlay? Relative position in parameter space


Image credit: Baerbel Koribalski


Summary


Summary
· Visualisation complements automated QC tests · Visualisation is the key to unlocking citizen science · We can leverage citizen science to find the weird (interesting) things, which we can analyse in detail
· WALLABY HI Analysis Tool (WHAT)