Galaxy Zoo Starburst Talk

Merging Classification Problem Fixed! Quench Boost 4 take 2.

  • mzevin1 by mzevin1 scientist, moderator


    As some of you have noticed, there have been some issues with the classification data on whether a galaxy is merging or not. Many galaxies that had clear signs of merging were being analyzed as 'Neither signatures merging or tidal debris'. This was not good - looking at mergers compared to non-mergers is very important for analyzing our post-quenched galaxies!

    It turns out that for many of these, the classifications were correct but the method in which we interpreted the classifications was wrong. Galaxies could either be classified as 'Merging', 'Tidal Debris', 'Both', or 'Neither'. One particular source, pointed out by zutopian, that showed clear signs of a merging event but was not included in the merging category is this galaxy. Out of 20 classifications, 3 classified it was merging, 5 as tidal debris, 5 as both, and 7 as neither. Since it had the most classifications as 'neither', it was classified as such. But wait! 13/20 (3 merging + 5 tidal + 5 both) said that in some way it had signatures of being a part of a merging event! Here's where the problem lie.

    To accommodate for this, a 5th category was added - 'Disturbed'. A galaxy is classified as 'disturbed' if the # of merging votes + # of tidal debris votes + # of both votes > # of neither votes. Galaxies are only classified as 'Neither' when # neither > # merging + # tidal debris + # both. This should fix the problem, and add a good deal more galaxies that have merger signatures to our data!

    Since we have many new galaxies classified as 'disturbed', it would be a great idea to retry the Quench Boost: A How-To-Guide, Part 4 with the new data! Now, when the 'neither' category is filtered out, it will still leave the galaxies in the new category of 'disturbed'. This should provide some interesting new results. Feel free to post anything you find on this thread!




  • JeanTate by JeanTate in response to mzevin1's comment.

    I've downloaded both the revised QS and QC catalogs, which I call v4.

    I have compared the values in each field, for each record, and found that only in the field 'merging' are there any changes.

    Altogether there are, for the QS v4 catalog, 107 records (objects, or 'galaxies') for which there is a change in the value in the field 'merging' (compared with the QS v3 catalog); for the v4 QC catalog, 83.


  • zutopian by zutopian

    Thanks for the updated data!

    I compared the previous QC data with the new QC data.:

    Neither: 2705 - 65 =2640

    Merging: 148 + 2 = 150

    Tidal debris: 127 +/-0 = 127

    Both: 13 +/-0= 13

    ◦[Star or artifact]: 9 -9= 0 (7 neither,1 merging,1 disturbed)

    null: 1 -1= 0 (1 merging)

    Disturbed 0 +73= 73

    3003 +75/-75 = 3003


  • JeanTate by JeanTate in response to JeanTate's comment.

    Of the 107 QS v3 -> v4 changes, 96 are 'Neither' -> 'Disturbed': AGS000000c, AGS000000q, AGS0000022, AGS000005r, AGS000008x, AGS000009b, AGS000009h, AGS000009p, AGS00000cg, AGS00000e0, AGS00000e5, AGS00000fq, AGS00000g3, AGS00000ge, AGS00000gg, AGS00000gl, AGS00000hn, AGS00000ib, AGS00000ix, AGS00000j5, AGS00000kg, AGS00000kh, AGS00000lp, AGS00000mn, AGS00000mx, AGS00000ny, AGS00000oq, AGS00000ox, AGS00000pd, AGS00000s5, AGS00000sa, AGS00000tt, AGS00000u4, AGS00000v9, AGS00000ve, AGS00000vq, AGS00000wi, AGS00000yk, AGS00000zx, AGS000010e, AGS000011p, AGS0000134, AGS000013p, AGS000013y, AGS0000142, AGS000014g, AGS000014l, AGS000016m, AGS000016v, AGS000018x, AGS00001ao, AGS00001c8, AGS00001cw, AGS00001ed, AGS00001ei, AGS00001fb, AGS00001g4, AGS00001gb, AGS00001gh, AGS00001h0, AGS00001iz, AGS00001jf, AGS00001jq, AGS00001ka, AGS00001kh, AGS00001li, AGS00001nt, AGS00001ou, AGS00001qe, AGS00001s8, AGS00001sn, AGS00001tk, AGS00001v1, AGS00001wg, AGS00001xi, AGS00001xq, AGS00001yq, AGS00001yx, AGS0000202, AGS0000203, AGS0000212, AGS0000214, AGS0000215, AGS000021u, AGS0000227, AGS000022o, AGS000022y, AGS000023d, AGS000023f, AGS0000261, AGS0000269, AGS0000272, AGS000027k, AGS000027u, AGS000028b, and AGS000029i.

    Eight are 'Star/artifact' objects, six of which are now classified as 'Neither' (in the 'merging' field), and two 'Tidal debris'. The six are: AGS000004v, AGS000005o, AGS00000l1, AGS00001uo, AGS0000265, and AGS00002ak; the two are: AGS00000uh and AGS000026g1. Thus all the 'star/artifact' objects have - in the v4 QS catalog - non-null values in the 'merging' field. Here are images of those two 'Tidal debris' star/artifact objects (respectively):

    enter image description here enter image description here

    There are also two 'Tidal debris' -> 'Neither' ( AGS000013g and AGS000020j); and one 'Neither' -> 'Tidal debris' ( AGS00001k4):

    enter image description here
    enter image description here
    enter image description here

    1 This is a most interesting object! It is discussed in the GZ forum, here; most likely it's an overlap with a Green Pea.


  • JeanTate by JeanTate

    Of the 83 QC v3 -> v4 changes, 72 are 'Neither' -> 'Disturbed': AGS00002c3, AGS00002cp, AGS00002d2, AGS00002dc, AGS00002eh, AGS00002f6, AGS00002g4, AGS00002hm, AGS00002mm, AGS00002mv, AGS00002qr, AGS00002qv, AGS00002r6, AGS00002to, AGS00002wb, AGS00002wd, AGS00002wg, AGS00002wk, AGS00002x4, AGS000030u, AGS000031i, AGS000031n, AGS000032n, AGS000035k, AGS000035n, AGS0000368, AGS000036o, AGS000036u, AGS000036y, AGS0000374, AGS000037o, AGS000038t, AGS00003e9, AGS00003he, AGS00003i6, AGS00003i8, AGS00003je, AGS00003l1, AGS00003mf, AGS00003mh, AGS00003mj, AGS00003ph, AGS00003q8, AGS00003ub, AGS00003vm, AGS00003wm, AGS00003xe, AGS00003xj, AGS00003y5, AGS00003yv, AGS00003z1, AGS0000405, AGS000040i, AGS000040q, AGS0000446, AGS000044x, AGS0000457, AGS000048y, AGS00004b1, AGS00004da, AGS00004ft, AGS00004gf, AGS00004gv, AGS00004h9, AGS00004hm, AGS00004ht, AGS00004ji, AGS00004la, AGS00004lj, AGS00004lm, AGS00004lz, and AGS00004nd.

    Nine are 'Star/artifact' objects, seven of which are now classified as 'Neither' (in the 'merging' field), one as 'Merging' ( AGS00003k3), and one as 'Disturbed ( AGS00002y7)1. Thus all the 'star/artifact' objects have - in the v4 QC catalog - non-null values in the 'merging' field. Here are images of the last two (respectively):

    enter image description here
    enter image description here

    The one other change is AGS00004n1, 'null' -> 'Merging'. This object should have been removed from the catalog (per Laura), as it is an incomplete duplicate of another ( AGS00004cy).

    I will compare my findings with those of zutopian in my next post.

    1 This is, of course, not a galaxy, but an HII region (or similar) in a very large galaxy (NGC 3198)


  • JeanTate by JeanTate in response to zutopian's comment.

    My numbers match zutopian's:

    • Disturbed 0 +73= 73: I get 72 ('Neither' -> 'Disturbed') plus 1 (star/artifact) = 73
    • Merging: 148 + 2 = 150: I get 148 (no change) plus 1 (star/artifact) plus 1 (null) = 150
    • [Star or artifact]: 9 -9= 0 (7 neither,1 merging,1 disturbed): I get the same
    • Neither: 2705 - 65 =2640: I get a change of -72 ('Neither' -> 'Disturbed') +7 (star/artifact -> 'Neither') = 65.

    And also no change to 'Both' or 'Tidal debris'.


  • zutopian by zutopian

    It is fine, that there are now more merger candidates available, but curiously not all of the missed mergers from my collection are classified as "disturbed". That's somehow disappointing and strange, because those are actually obvious ones in my opinion.:
    My collection contains 15 missed mergers, but just 6 are now classified as #disturbed: AGS00000wi , AGS000010e , AGS0000227 , AGS00001iz , AGS00003xe , AGS0000022. 9 are still classified as "neither"! Could you please check, if those are mergers and please look at the votes!


  • JeanTate by JeanTate in response to mzevin1's comment.

    While the revision does indeed boost the number of galaxies with classifications as != 'Neither', in both the QS and QC catalogs, I'm not sure it's all that much of an improvement.

    You see, the 'mistake' object ( AGS00004n1) in QC is still in the catalog, and the one it's a (partial) duplicate of ( AGS00004cy) has not been edited (it's not clear which fields refer to DR7 ObjId 587732591182020753 and which refer to some other object). Further, the sdss_id field still contains the 'last few digits mangled' values, not the actual DR7 ObjIds. And none of the outliers clearly identified some time ago have been removed1. Nor have any of the outliers in the v3 catalogs been explained, much less removed. And the other changes in classifications - from v2 to v3 - have yet been explained.

    If there were only a few outliers and apparently anomalous entries (objects), I guess it wouldn't matter much, in terms of doing Tools-based analyses in which the 'merging fraction' is a key variable; after all, there are now ~350 '!=Neither' QC objects. But the number of potentially problematic objects is now in the dozens.

    Other than to gain experience in using Tools to do analyses, what's the point of spending time and effort, when we know for certain that the two catalogs are not 'clean'?

    1 Yes, the QS-QC duplicates are gone (they were removed in the v3 catalog), as well as all the QC-QC ones. Also, one QC object in the v2 catalog that has an unreliable redshift was removed in the v3 one.


  • mlpeck by mlpeck in response to JeanTate's comment.

    Here's the skyserver Object explorer page for object AGS00004cy: Note that its position is different from its entry in the QC data table while it's given correctly in its Quench Talk page and in GZ examine.

    The photometry and spectroscopic quantities for the object with uid AGS00004cy in the data table appear to apply to AGS00004n1, which is the object that's actually at the listed position:

    I don't know about the classifications -- I wouldn't call that object completely round, but that doesn't mean a majority of GZ classifiers couldn't have seen it as round.


  • JeanTate by JeanTate in response to mlpeck's comment.

    There's a couple of posts in this thread on these two 'objects' ( AGS00004n1, and AGS00004cy). I think you have, independently, confirmed what was written there! 😃


  • mlpeck by mlpeck

    Did I? I had a little trouble following the edits in that thread, and the switch from "raw" model magnitudes to k-corrected (we hope) ones makes comparing the photometry a little difficult.

    In any case, line 2596 in the current data table has the correct spectroscopic data for the object that's actually at the listed position, which is AGS00004n1. The photometry could plausibly be k-corrected magnitudes for the same object I suppose. The uid and image fields in that line are incorrect.


  • zutopian by zutopian

    Referring to mzevin's statement:

    "Since we have many new galaxies classified as 'disturbed', it would be a great idea to retry the Quench Boost: A How-To-Guide, Part 4 with the new data! Now, when the 'neither' category is filtered out, it will still leave the galaxies in the new category of 'disturbed'. This should provide some interesting new results. Feel free to post anything you find on this thread!"

    mzevin's comment in another topic (before the update: on 17th Sept)

    "We'll definitely have to check out if the amount of wrong classifications of mergers and nonmergers will be statistically significant. I'm guessing that most of the ones wrongly classified as mergers had foreground stars that people believed were galaxies. Depending on the frequency of this we may have to go through the galaxies classified as mergers and double check with the spectral data to confirm - looking at merger signatures is really important to the science goals of this project, and we have to make sure that these classifications (for the most part) are correct!"

    My comment:
    There are still those wrong merger classifications (and however also missed mergers) and the quantity of wrong mergers increased after the update. So I think, that it doesn't make sense to use the updated data for merger analysis, because the analysis results would be wrong!
    EDIT: BTW, I wonder, if the current classification results (merger and the other classifications) are already "bias-corrected"?


  • mzevin1 by mzevin1 scientist, moderator

    The update did not help to eliminate foreground stars that were classified as 'mergers', but it should have increased the number of actual mergers, albeit a few false mergers probably snuck in in the process. So, I'm hoping that the number of new true mergers outweighs the number of new false mergers with the update.


  • zutopian by zutopian

    I noticed, that there is another change in the QS catalogue (downloaded table), but it isn't concerning classification data.
    In the previous QS catalog there was given however sometimes "null, null, null, null" e.g. AGS0000004, but now there is given a value e.g.1.4834999.
    I wonder why?


  • JeanTate by JeanTate in response to zutopian's comment.

    The field "sfr" was added in v3 of the QS catalog, and is also in v4 (with an unchanged value).


  • zutopian by zutopian

    After the update following missed merger is however still "neither".:

    I found out, that it is presented in below paper.: So I think, that the QS catalogue must be corrected so, that it is merger.: AGS0000172

    E+A and Companion Galaxies - I : A Catalogue and Statistics

    Based on our intensive spectroscopic campaign with the GoldCam spectrograph on the Kitt Peak National Observatory (KPNO) 2.1-m telescope, we have constructed the first catalogue of E+A galaxies with spectroscopic companion galaxies

    Chisato Yamauchi (ISAS/Jaxa), Masafumi Yagi (NAOJ), Tomotsugu Goto (Ifa/Uh)

    (Submitted on 4 Sep 2008)

    PS: In Fig. 4 there are shown images.


  • zutopian by zutopian

    Here is the link to the paper, which I mentioned in my previous post.:

    PS : However it doesn't work to edit that post, but I had edited posts before. When I click EDIT, nothing happens.
    EDIT: PPS: To edit this post worked. 😃


  • zutopian by zutopian

    I should check, if the QS sample contains more GPairs from that paper. Or maybe someone else could please do this?
    Besides the QC sample should also be checked.

    Here is the table of companion pairs.:
    It is also given in the arxiv version on page 5.:

    Besides the rejected candidates (=non-companion) should also be checked. I couldn't find a related table.:

    We observed 26 companion candidates of E+A galaxies at the KPNO to measure their redshifts. Their spectra showed that 17 targets are true companion galaxies. The number of spectroscopically-confirmed E+A’s companions are now 34.