pith. sign in

arxiv: 2606.24807 · v1 · pith:TYHEI4CYnew · submitted 2026-06-23 · 🌌 astro-ph.IM

The Impact of Host Galaxy Properties on Supernova Classification with Hierarchical Labels

V. Ashley Villar , Sebastian Gomez , Edo Berger , Alex Gagliano This is my paper

Pith reviewed 2026-06-25 22:35 UTC · model grok-4.3

classification 🌌 astro-ph.IM
keywords supernova classificationhost galaxy propertiesphotometric classificationhierarchical labelsType Ia supernovaeType II supernovaesuperluminous supernovaeweighted hierarchical cross-entropy
0
0 comments X

The pith

Host galaxy properties alone can isolate over 90 percent pure samples of Type Ia supernovae with or without redshift information.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether observable properties of a supernova's host galaxy improve photometric classification of supernova types. It demonstrates that these properties by themselves produce samples of Type Ia events with purity above 90 percent. Adding redshift information further allows somewhat pure samples of Type II and superluminous supernovae. A new weighted hierarchical cross-entropy loss is used to train the classifier in a way that respects the natural hierarchy of transient classes. This matters for upcoming surveys that will detect far more supernovae than can receive spectroscopic follow-up.

Core claim

Host galaxy information alone successfully isolates relatively pure samples of Type Ia supernovae exceeding 90 percent purity, with or without redshift. With redshift, samples of Type II supernovae and superluminous supernovae exceeding 70 percent purity can also be obtained. Host galaxy properties do not significantly improve classification accuracy when complete light curves and redshifts are available, but they do when redshift is absent. A new objective function, the weighted hierarchical cross-entropy, is applied for the first time to supernova classification, and new classifications are provided for the Pan-STARRS Medium Deep Survey photometric sample.

What carries the argument

The weighted hierarchical cross-entropy objective function, which accounts for the hierarchical structure of supernova classes when training photometric classifiers that use host galaxy properties.

If this is right

  • The full photometric sample of Pan-STARRS Medium Deep Survey supernovae can be increased to more than 4400 events.
  • Real-time selection of events for spectroscopic followup can use host galaxy properties to achieve high purity without redshift or light-curve data.
  • Archival studies can draw subsamples of Type Ia, Type II, and superluminous supernovae at stated purity levels from photometry alone.
  • Classification performance improves measurably when redshift is unavailable but host galaxy data are included.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same host-galaxy approach could be tested on other classes of transients whose environments differ systematically.
  • Large future surveys may incorporate existing galaxy catalogs as a first-stage filter before light-curve processing.
  • If the representativeness assumption holds, host properties appear to encode environmental information that is partially independent of light-curve shape.

Load-bearing premise

The spectroscopic training sample is representative of the photometric target population in host galaxy property distributions and the assigned hierarchical class labels contain no systematic bias.

What would settle it

Training the classifier on the spectroscopic sample and then measuring purity on a new set of events that later receive spectroscopic labels, finding that Type Ia purity falls below 90 percent.

Figures

Figures reproduced from arXiv: 2606.24807 by Alex Gagliano, Edo Berger, Sebastian Gomez, V. Ashley Villar.

Figure 1
Figure 1. Figure 1: Scatter plot of various host galaxy features to classify SNe in this work (see Sec. 2 for feature definitions). SNe are spectroscopically labelled as part of PS1 MDS, and for features where the host galaxy features were not observed, they are inferred via K-means imputation. Even in these simple feature spaces, clear clustering is seen. Many host galaxy features are available via FLEET, including multi-sur… view at source ↗
Figure 2
Figure 2. Figure 2: Hierarchical graph structure used in this work. Blue text is meant as a guide to the various components of a generic tree. p(Ib/c|H-poor) = p(Ib/c) p(H-poor), (7) where p(H-poor) is the sum of the probabilities of all H-poor descendant classes. Analogous to Eqn.3, we can define the weighted hier￾archical cross-entropy (WHXE) as: LWHXE(c (h) i ) = − H X−1 h=h′ W(c (h) )λ(c (h) ) log p(c (h) j |c (h+1) k ), … view at source ↗
Figure 3
Figure 3. Figure 3: Purity (top row), completeness (middle row) and F1-score (bottom row) achieved with the various feature sets and confidence thresholds as a function of SN type (columns). Shaded regions represent 1σ uncertainties, computed using the same classifier with different random model initializations during training [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Confusion matrices for the classifier using only host galaxy information without (left) and with (right) redshift information. Numbers are overall percent, while uncertainties are given in parentheses (e.g., the completeness of SLSNe in the five-way classifier is 0.48 ± 0.06). The classifier reaches state-of-the-art results for Type Ia vs CC classification, but fails to achieve high accuracy in the five-wa… view at source ↗
Figure 5
Figure 5. Figure 5: Confusion matrices for the classifier using solely light curve features (left) and a combination of galaxy and light curve features (right), both excluding redshift information. Inclusion of galaxy features does not improve classification for any one class to a statistically significant degree, and the overall performance only moderately increases. F1-score of 0.48. Without redshift information, we achieve… view at source ↗
Figure 6
Figure 6. Figure 6: Confusion matrices for the classifier using redshift information, trained using only light curve features (left) and with both light curve and galaxy features (right). Including host information provides similar overall classification accuracy. of LSST (Graham et al. 2018). Transients which are primarily found in intrinsically low-luminosity galaxies (e.g., SLSNe) will have particularly unreliable redshift… view at source ↗
Figure 7
Figure 7. Figure 7: Breakdown of SN subclasses in our new photometrically classified set of SNe without redshift information. Compared to the spectroscopic sample (green), our classifier (blue) predicts a smaller fraction of Type Ia and Type II SNe and a larger fraction of SLSNe and Type Ibc SNe. However, using our purity matrix (see text), we can correct our classification breakdowns accounting for expected mis-identificatio… view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of SN classifications from Villar et al. (2020) and this work. On the abscissa, we plot the confidence level of the Villar et al. (2018) classifier, while the ordinate shows the confidence of the new classifier for the same class. Objects plotted as a point have the same classification with both methods; objects plotted as an ‘x’ have different classi￾fications with either method. The color of t… view at source ↗
read the original abstract

With the advent of the Vera C. Rubin Observatory, the discovery rate of supernovae (SNe) will surpass the rate of SNe with real time spectroscopic followup by three orders of magnitude. Accurate photometric classifiers are essential to both select interest events for followup in real time and for archival population-level studies. In this work, we investigate the impact of observable host galaxy information on the classification of SNe, both with and without additional light curve and redshift information. We find that host galaxy information alone can successfully isolate relatively pure (>90%) samples of Type Ia SNe with or without redshift information. With redshift information, we can additionally produce somewhat pure (>70%) samples of Type II SNe and superluminous supernovae. Additionally with redshift information, host galaxy properties do not significantly improve the accuracy of SN classification when paired with complete light curves. In the absence of redshift information, however, galaxy properties significantly increase the accuracy of photometric classification. As a part of this analysis, we present the first formal application of a new objective function, the weighted hierarchical cross-entropy, to the problem of supernova classification. This objective function more naturally accounts for the hierarchical nature of supernova classes and, more broadly, transients. Finally, we present a new set of SN classifications for the Pan-STARRS Medium Deep Survey of SNe that lack spectroscopic redshift, increasing the full photometric sample to >4400 events.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper examines the utility of host-galaxy observables (stellar mass, SFR, color, etc.) for photometric supernova classification, both alone and combined with light curves or redshift. It reports that host properties alone yield >90% pure Type Ia samples (with or without redshift) and, with redshift, >70% pure Type II and SLSN samples. The authors introduce a weighted hierarchical cross-entropy loss to respect the SN taxonomy hierarchy, apply the classifier to the Pan-STARRS Medium Deep Survey, and release classifications for >4400 photometrically observed events lacking spectroscopic redshifts.

Significance. If the reported purities are robust, the work directly supports real-time transient selection and archival studies for Rubin/LSST, where spectroscopic follow-up will be scarce. The weighted hierarchical cross-entropy is presented as the first formal application to SN classification and offers a more natural treatment of hierarchical labels than flat cross-entropy; this methodological contribution is reusable for other transient taxonomies. The release of >4400 new classifications also augments the public Pan-STARRS sample.

major comments (3)
  1. [Abstract and Methods] Abstract and Methods: concrete purity thresholds (>90% Ia, >70% II/SLSN) are stated without accompanying training-set size, photometric test-set size, cross-validation procedure, or uncertainty estimates on the purity metrics. These omissions leave the central quantitative claims without visible statistical support.
  2. [Methods] Methods (domain-shift discussion): no quantitative comparison (KS test, propensity-score matching, or re-weighting) is reported between the joint distribution of host-galaxy properties in the spectroscopically labeled training sample and the unlabeled Pan-STARRS photometric population. Because spectroscopic follow-up is known to favor brighter hosts, any mismatch directly affects the learned decision boundaries of the weighted hierarchical cross-entropy classifier and renders the reported purities dependent on an untested transfer assumption.
  3. [Results] Results: the claim that host properties “significantly increase” accuracy when redshift is absent is presented without a direct ablation (light-curve-only vs. light-curve+host) on the same test objects or with error bars, making it impossible to judge the magnitude or statistical significance of the improvement.
minor comments (2)
  1. [Methods] Notation for the hierarchical loss weights is introduced without an explicit equation or table listing the numerical values used in the reported experiments.
  2. [Figures] Figure captions for the purity/completeness curves do not state the exact number of objects or the train/test split underlying each curve.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their detailed and constructive report. We address each major comment below and have revised the manuscript accordingly to strengthen the statistical support and transparency of our results.

read point-by-point responses

  1. Referee: [Abstract and Methods] Abstract and Methods: concrete purity thresholds (>90% Ia, >70% II/SLSN) are stated without accompanying training-set size, photometric test-set size, cross-validation procedure, or uncertainty estimates on the purity metrics. These omissions leave the central quantitative claims without visible statistical support.

    Authors: We agree that these supporting details are necessary for evaluating the reported purities. The revised manuscript updates the abstract and methods section to explicitly state the training-set size (spectroscopically confirmed events used for model fitting), the photometric test-set size drawn from Pan-STARRS, the cross-validation procedure employed, and bootstrap-derived uncertainty estimates on all purity metrics. revision: yes

  2. Referee: [Methods] Methods (domain-shift discussion): no quantitative comparison (KS test, propensity-score matching, or re-weighting) is reported between the joint distribution of host-galaxy properties in the spectroscopically labeled training sample and the unlabeled Pan-STARRS photometric population. Because spectroscopic follow-up is known to favor brighter hosts, any mismatch directly affects the learned decision boundaries of the weighted hierarchical cross-entropy classifier and renders the reported purities dependent on an untested transfer assumption.

    Authors: This point is well taken. The revised methods section now includes a quantitative domain-shift analysis using two-sample Kolmogorov-Smirnov tests on the marginal and joint distributions of host-galaxy stellar mass, star-formation rate, and color between the spectroscopic training sample and the Pan-STARRS photometric population. We also discuss the implications of any detected differences for the transfer of the classifier. revision: yes

  3. Referee: [Results] Results: the claim that host properties “significantly increase” accuracy when redshift is absent is presented without a direct ablation (light-curve-only vs. light-curve+host) on the same test objects or with error bars, making it impossible to judge the magnitude or statistical significance of the improvement.

    Authors: We acknowledge the need for a controlled ablation study. The revised results section now presents a direct comparison of light-curve-only versus light-curve-plus-host models evaluated on identical test objects, with all metrics accompanied by error bars obtained from the cross-validation procedure. This allows quantitative assessment of the improvement and its statistical significance. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical classification results on survey data

full rationale

The paper trains a classifier (using weighted hierarchical cross-entropy loss) on spectroscopically labeled supernovae and evaluates purity on the Pan-STARRS photometric sample. Reported accuracies and purities are direct outputs of this train/test procedure on real data; no step reduces a claimed prediction to a fitted input by construction, nor does any result depend on a self-citation chain or ansatz smuggled from prior work. The analysis is self-contained against external survey benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Central claims rest on standard supervised-learning assumptions plus tunable weights inside the new objective function; no new physical entities are introduced.

free parameters (1)
  • class and hierarchy weights in weighted hierarchical cross-entropy
    The objective function requires weights that balance contributions across class levels and are chosen to optimize performance on the training set.
axioms (1)
  • domain assumption Spectroscopically confirmed supernovae form an unbiased training distribution for photometric events with respect to host galaxy properties.
    Invoked when models trained on labeled events are applied to unlabeled photometric samples.

pith-pipeline@v0.9.1-grok · 5788 in / 1353 out tokens · 41100 ms · 2026-06-25T22:35:23.025650+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 6 canonical work pages · 2 internal anchors

  1. [1]

    2020, Astron

    Aghanim, N., et al. 2020, Astron. Astrophys, 641, A6

    2020

  2. [2]

    , archivePrefix = "arXiv", eprint =

    Anderson, J. P., Habergham, S., James, P., & Hamuy, M. 2012, Monthly Notices of the Royal Astronomical Society, 424, 1372 Astropy Collaboration, Price-Whelan, A. M., Sip˝ ocz, B. M., et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ, 935, 167, doi: 10.3847/1538-4357/ac7c74

    work page doi:10.3847/1538-3881/aabc4f 2012

  3. [3]

    2020, The Astrophysical Journal, 902, 60

    Coppejans, D. 2020, The Astrophysical Journal, 902, 60

    2020

  4. [4]

    2010, The Astrophysical Journal, 722, 1946

    Berger, E. 2010, The Astrophysical Journal, 722, 1946

    2010

  5. [5]

    Lord, N. A. 2020, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 12506–12515

    2020

  6. [6]

    S., Kulkarni, S

    Bloom, J. S., Kulkarni, S. R., & Djorgovski, S. G. 2002, The Astronomical Journal, 123, 1111

    2002

  7. [7]

    2021, The Astronomical Journal, 162, 275

    Boone, K. 2021, The Astronomical Journal, 162, 275

    2021

  8. [8]

    J., Gal-Yam, A., Schulze, S., et al

    Bruch, R. J., Gal-Yam, A., Schulze, S., et al. 2021, The Astrophysical Journal, 912, 46

    2021

  9. [9]

    2021, The Astronomical Journal, 162, 231

    Carrasco-Davis, R., Reyes, E., Valenzuela, C., et al. 2021, The Astronomical Journal, 162, 231

    2021

  10. [10]

    2013, The Astrophysical Journal, 770, 107 Hierarchical SN Classification17

    Childress, M., Aldering, G., Antilogus, P., et al. 2013, The Astrophysical Journal, 770, 107 Hierarchical SN Classification17

    2013

  11. [11]

    2022, The Astrophysical Journal, 927, 199

    Dong, Y., Milisavljevic, D., Leja, J., et al. 2022, The Astrophysical Journal, 927, 199

    2022

  12. [12]

    J., & Mandel, K

    Foley, R. J., & Mandel, K. 2013, The Astrophysical Journal, 778, 167

    2013

  13. [13]

    Aleo, P. D. 2023, arXiv preprint arXiv:2305.08894

    arXiv 2023

  14. [14]

    2021, The Astrophysical Journal, 908, 170

    Gagliano, A., Narayan, G., Engel, A., et al. 2021, The Astrophysical Journal, 908, 170

    2021

  15. [15]

    Observational and Physical Classification of Supernovae

    Gal-Yam, A. 2017, arXiv preprint arXiv:1611.09353, 195, doi: 10.1007/978-3-319-21846-5 35

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1007/978-3-319-21846-5 2017

  16. [16]

    K., et al

    Gomez, S., Berger, E., Blanchard, P. K., et al. 2023, The Astrophysical Journal, 949, 114 —. 2020, The Astrophysical Journal, 904, 74

    2023

  17. [17]

    L., Connolly, A

    Graham, M. L., Connolly, A. J., Ivezi´ c,ˇZ., et al. 2018, AJ, 155, 1, doi: 10.3847/1538-3881/aa99d4

    work page doi:10.3847/1538-3881/aa99d4 2018

  18. [18]

    2012, Monthly Notices of the Royal Astronomical Society, 424, 2841

    Habergham, S., James, P., & Anderson, J. 2012, Monthly Notices of the Royal Astronomical Society, 424, 2841

    2012

  19. [19]

    Z., Aramyan, L., et al

    Hakobyan, A., Adibekyan, V. Z., Aramyan, L., et al. 2012, Astronomy & Astrophysics, 544, A81

    2012

  20. [20]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2

    work page doi:10.1038/s41586-020-2649-2 2020

  21. [21]

    A., et al

    Hosseinzadeh, G., Dauphin, F., Villar, V. A., et al. 2020, The Astrophysical Journal, 905, 93

    2020

  22. [22]

    A., & Berger, E

    Hsu, B., Hosseinzadeh, G., Villar, V. A., & Berger, E. 2022, arXiv preprint arXiv:2204.09809

    arXiv 2022

  23. [23]

    2012, VizieR Online Data Catalog, II

    Hudelot, P., Cuillandre, J.-C., Withington, K., et al. 2012, VizieR Online Data Catalog, II

    2012

  24. [24]

    Hunter, J. D. 2007, Computing in Science & Engineering, 9, 90, doi: 10.1109/MCSE.2007.55 Jacobson-Gal´ an, W., Dessart, L., Davis, K., et al. 2024, arXiv preprint arXiv:2403.02382

    work page doi:10.1109/mcse.2007.55 2007

  25. [25]

    M., Kulkarni, S., Gal-Yam, A., et al

    Kasliwal, M. M., Kulkarni, S., Gal-Yam, A., et al. 2012, The Astrophysical Journal, 755, 161

    2012

  26. [26]

    L., & Kirshner, R

    Kelly, P. L., & Kirshner, R. P. 2012, The Astrophysical Journal, 759, 107

    2012

  27. [27]

    2019, Publications of the Astronomical Society of the Pacific, 131, 094501

    Kessler, R., Narayan, G., Avelino, A., et al. 2019, Publications of the Astronomical Society of the Pacific, 131, 094501

    2019

  28. [28]

    2016, The Astrophysical Journal, 818, 3

    Khazov, D., Yaron, O., Gal-Yam, A., et al. 2016, The Astrophysical Journal, 818, 3

    2016

  29. [29]

    P., & Ba, J

    Kingma, D. P., & Ba, J. 2014, arXiv preprint arXiv:1412.6980

    Pith/arXiv arXiv 2014

  30. [30]

    2022, arXiv e-prints, arXiv:2209.02784

    Kisley, M., Qin, Y.-J., Zabludoff, A., Barnard, K., & Ko, C.-L. 2022, arXiv e-prints, arXiv:2209.02784. https://arxiv.org/abs/2209.02784

    arXiv 2022

  31. [31]

    A., & Hlozek, R

    Kunz, M., Bassett, B. A., & Hlozek, R. A. 2007, Physical Review D—Particles, Fields, Gravitation, and Cosmology, 75, 103508

    2007

  32. [32]

    Leaman, J., Li, W., Chornock, R., & Filippenko, A. V. 2011, Monthly Notices of the Royal Astronomical Society, 412, 1419 M¨ oller, A., & de Boissi` ere, T. 2020, Monthly Notices of the Royal Astronomical Society, 491, 4277

    2011

  33. [33]

    S., Biswas, R., & Hloˇ zek, R

    Muthukrishna, D., Narayan, G., Mandel, K. S., Biswas, R., & Hloˇ zek, R. 2019, Publications of the Astronomical Society of the Pacific, 131, 118002

    2019

  34. [34]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library

    Paszke, A., Gross, S., Massa, F., et al. 2019, arXiv e-prints, arXiv:1912.01703, doi: 10.48550/arXiv.1912.01703

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1912.01703 2019

  35. [35]

    2022, The Astrophysical Journal Supplement Series, 259, 13

    Qin, Y.-J., Zabludoff, A., Kisley, M., et al. 2022, The Astrophysical Journal Supplement Series, 259, 13

    2022

  36. [36]

    2022, The Astronomical Journal, 163, 57

    Qu, H., & Sako, M. 2022, The Astronomical Journal, 163, 57

    2022

  37. [37]

    A., & Percival, S

    James, P. A., & Percival, S. M. 2022, Monthly Notices of the Royal Astronomical Society, 513, 3564

    2022

  38. [38]

    P., Tollerud, E

    Robitaille, T. P., Tollerud, E. J., Greenfield, P., et al. 2013, Astronomy & Astrophysics, 558, A33 S´ anchez-S´ aez, P., Reyes, I., Valenzuela, C., et al. 2021, The Astronomical Journal, 161, 141

    2013

  39. [39]

    2021, The Astrophysical Journal Supplement Series, 255, 29

    Schulze, S., Yaron, O., Sollerman, J., et al. 2021, The Astrophysical Journal Supplement Series, 255, 29

    2021

  40. [40]

    2019, arXiv preprint arXiv:1903.02476

    Swann, E., Sullivan, M., Carrick, J., et al. 2019, arXiv preprint arXiv:1903.02476

    Pith/arXiv arXiv 2019

  41. [41]

    2019, The Astrophysical Journal, 884, 83

    Villar, V., Berger, E., Miller, G., et al. 2019, The Astrophysical Journal, 884, 83

    2019

  42. [42]

    A., de Soto, K., & Gagliano, A

    Villar, V. A., de Soto, K., & Gagliano, A. 2023, arXiv preprint arXiv:2312.02266

    arXiv 2023

  43. [43]

    A., Nicholl, M., & Berger, E

    Villar, V. A., Nicholl, M., & Berger, E. 2018, The Astrophysical Journal, 869, 166

    2018

  44. [44]

    A., Hosseinzadeh, G., Berger, E., et al

    Villar, V. A., Hosseinzadeh, G., Berger, E., et al. 2020, The Astrophysical Journal, 905, 94

    2020