PRISM-Astro/h2f_ps1_pasquet_autolabeling_radius_norm_p6

Description

This is the spectroscopic galaxy sample of Pasquet et al. [1], originally built for photometric redshift estimation from SDSS, re-imaged as multi-resolution PanSTARRS cutouts through hips2fits (PanSTARRS DR1 HiPS, served by CDS).

It contains 639,750 examples drawn from 481,589 galaxies, built by autolabeling: the cutouts are not centered on the galaxy but on a simulated transient position. For each galaxy, positions are drawn uniformly inside the elliptical footprint of its r-band Sersic fit, truncated at 3·Re.

This is exactly how h2f_ps1_pasquet_autolabeling_0_01pct was built. The only difference is how many positions each galaxy gets, and that is what the radius_norm name refers to.

Every cutout is centered on the simulated transient position, and the host_pos label is the offset from there to the galaxy. Because the redshift of the galaxy travels with each example, the dataset serves both host identification and redshift estimation.

How many examples per galaxy: radius_norm

The problem

A galaxy can only place a transient as far as 3·Re from its own center. A small galaxy simply cannot produce a large transient-host distance.

Most galaxies are small. So if every galaxy contributes roughly the same number of examples, nearly all of them land at short distances and the long ones are almost missing. In the 0.01pct version, the fullest distance bin holds 9,487 times more examples than the emptiest one.

The rule

Bigger galaxies generate more examples:

n = max(1, round((Re / Re0) ** p))        Re0 = 8 arcsec,  p = 6

Re0 is the size at which a galaxy starts generating more than one example. Below it, a galaxy contributes exactly one. No galaxy is ever dropped: all 481,589 are present, and 98.5 % of them contribute a single example. Only 7,415 galaxies (1.5 %) contribute more, up to 911 for the largest one.

Why those numbers

The count of galaxies large enough to reach a given distance drops very fast: 143,496 reach 10 arcsec, 3,968 reach 30, and only 129 reach 60. Raising the count to a power compensates that drop, and p = 6 with Re0 = 8 is the pair that flattened the histogram best in a grid search over both.

They are not universal constants: they were fitted to the size distribution of this catalog and must be recomputed for a different survey. The companion DELIGHT datasets use p = 6, Re0 = 12 and p = 4, Re0 = 8.

Result

	0.01pct	radius_norm_p6
examples	506,572	639,750
examples per galaxy	1.05 avg	1.33 avg
fullest / emptiest distance bin	9,487×	93×

Measured over 5 linear bins between 0″ and 70″, the grid used to report per-distance metrics. Neither dataset leaves an empty bin.

How it was built

491,537 rows are rows of 0.01pct, copied unchanged. Because both use the same uniform-in-area law, keeping a random subset of each galaxy's rows is a valid sample of that same law, so only the shortfall had to be downloaded: 148,213 new cutouts, all from the 5,839 galaxies that did not already have enough. 15,035 surplus rows were dropped.

The new positions were generated with the same uniform-in-area law and their cutouts downloaded from hips2fits with the same pipeline, so the two halves are statistically indistinguishable. Every downloaded image was checked: no NaNs, no fully blank bands, and only 1.6 % of them contain any exactly-zero pixel (0.001 % of all pixels), which is the usual PanSTARRS mosaic coverage gap.

Cross-validation

The fold column holds a 5-fold split stratified by spectroscopic redshift. Redshift is discretized into 180 uniform bins over [0, 0.4], the same binning PRISM uses for its classification head, and the split balances those classes across folds. The 24 galaxies falling in bins with fewer than 5 members carry fold = 0 and belong to training only, since they cannot be stratified.

Note that folds are balanced by galaxy count, not by galaxy size, and Re**6 amplifies any size asymmetry between them.

Image format

image has shape (639750, 5, 5, 30, 30), which maps to (examples, resolution levels, photometric bands, height, width). Bands are grizy in that order. Level L has a pixel scale of 0.25 * 2^L arcsec/px, so level 0 is the finest and level 4 covers the widest field.

Floating point columns are stored as float32. For ra and dec this caps angular precision at roughly 0.08 arcsec; use objID for exact crossmatching.

Columns

Column	Type	Description
Image and label
image	(5,5,30,30) float32	Multi-resolution PanSTARRS grizy cutout, centered on the simulated transient position.
host_pos	float32 × 2	Offset [x, y] from the simulated transient to the galaxy, in level-0 pixels.
Identification
objID	int64	SDSS photometric object identifier. Repeats across the draws of the same galaxy, and is the key to group them.
specObjID	int64	SDSS spectroscopic identifier of the spectrum the redshift comes from. Unique per galaxy.
bestObjID	int64	Photometric object SDSS considers the best match to that spectrum.
ps1_objID	int64	PanSTARRS DR1 identifier of the crossmatched source.
sep_arcsec	float32	Angular separation of the SDSS–PanSTARRS crossmatch, in arcsec. At most 1.35.
dup_ambiguo	int64	1 when several SDSS objects matched the same PanSTARRS source at a comparable distance, 0 otherwise. Set for 59 galaxies.
Sky position
ra	float32	Right ascension of the galaxy, in degrees. Not the center of this cutout.
dec	float32	Declination of the galaxy, in degrees. Not the center of this cutout.
Redshift
z_spec	float32	Spectroscopic redshift, the regression target. Ranges from 0 to 0.4, with a median of 0.102.
z_spec_err	float32	Uncertainty reported by the SDSS spectroscopic pipeline.
z_class	int64	Redshift bin, 0 to 179, over [0, 0.4]. Used to stratify the folds and as the classification target.
Photometry and extinction
petroR90_r	float32	Radius enclosing 90% of the Petrosian flux in the r band, in arcsec.
dered_petro_r	float32	Dereddened r-band Petrosian magnitude.
extinction_r	float32	Galactic extinction in the r band, in magnitudes.
ebv_sfd	float32	Reddening E(B−V) from the SFD dust map.
ebv_from_ext_r	float32	Reddening E(B−V) implied by extinction_r, kept as a consistency check on the previous column.
Sérsic fit — r band, from CasJobs
rSerRadius	float32	Effective radius Re of the r-band Sérsic fit, in arcsec. Defines the sampling region.
rSerAb	float32	Axis ratio b/a of the fitted ellipse, between 0 and 1. Defines the sampling region.
rSerPhi	float32	Position angle of the fitted ellipse, in degrees. Defines the sampling region.
Cross-validation
fold	int64	Validation fold, 1 to 5, stratified by z_class. A value of 0 means the galaxy is used for training in every fold.

References

[1] Pasquet, J., Bertin, E., Treyer, M., Arnouts, S., & Fouchez, D. 2019, Photometric redshifts from SDSS images using a convolutional neural network, Astronomy & Astrophysics, 621, A26. doi:10.1051/0004-6361/201833617

@article{Pasquet_2019,
  doi = {10.1051/0004-6361/201833617},
  url = {https://doi.org/10.1051/0004-6361/201833617},
  year = {2019},
  month = {jan},
  publisher = {EDP Sciences},
  volume = {621},
  pages = {A26},
  author = {Pasquet, J. and Bertin, E. and Treyer, M. and Arnouts, S. and Fouchez, D.},
  title = {Photometric redshifts from SDSS images using a convolutional neural network},
  journal = {Astronomy \& Astrophysics},
}