Update: I'm closing the current Vasco project.
https://github.com/jannefi/vasco/ - because I've already started implementing a new one "Vasco60" (name comes from new tile size: 60x60")
My last "Vasco30" stats after a full run. Tiles:
11,774. Potential sources after source extractor pass 2:
18,38 million (this includes some duplicates). Reductions: Gaia and PS1 within 5", MNRAS 2022 specified filters, tiles near plate edges, and duplicate removal: R=71,624. Post-pipeline reductions:
- Stage 0: final PS1 veto (bug in pipeline caused problems to early PS1 veto), removal of tiles in Southern Hemisphere. R = 14,5K
- Stage 1: Skybot. Reduction was few hundred, but the whole run took 36 hours...
- Stage 2: SuperCosmos. Again a minor reduction
- Stage 3: PTF, around 700 matches, R = roughly 12K
- Stage 4: VSX, minor reduction again, about 100
Final R =
11,848
I did experimental reduction rounds with both USNO-B and locally mirrored MAPS catalog. This step would drop about 4K rows, but I have no clue if I should even report this
It's clear that main reductions come from two optical catalogs: PS1 and Gaia. Rest of the steps are needed, but their impact on R is small.
Few days ago I was desperately trying to compare my R to MNRAS 2022 R (5,399). I've now understood that these two numbers cannot be, and should not be compared. MNRAS 2022 R is full of unknowns, meaning it is not fully clear how exactly they ended up with that dataset. This was one of the topics discussed by Watters et al, and also Villarroel et al in their response. I will simply stop comparing. My small dataset does have around 170 tiles with matching coordinates in MNRAS 2022 R (with some duplicates), in total little bit over 100 matching coordinates. And much more in the bigger dataset they published.
I took some time and checked few coordinates in MNRAS R set that
did not end up in my R. I wanted to understand why they were not in my dataset: I had a corresponding tile. I downloaded separately a 60x60 arcmin tile with center coords: RA = 130.013°, Dec = +33.081° and started going through each pipeline step. All 7 candidate coordinates can be found from that tile. In the raw source extractor pass2 catalogue for this tile, 6/7 R points have a detection within 5″, one was "cut" by pass 1 most likely. It was bit surprising to see what cut the remaining six: the quality gates defined in the MNRAS 2022. Most of them failed several gates. So how come they are in
MNRAS 2022 R? I don't know. I made a full description of test steps, quality gate failure matrix and sent that to some of my "paper contacts", but crucially one of them has not replied to any of my e-mails. Lead author has been apparently too busy with other affairs. Note: this was a very small dataset so all this is speculative - except the reason why those six candidates didn't pass my pipeline.
Like said, I have started implementing Vasco60. Repository exist but it's currently private. Code is about 95% complete. I will make it public once it's ready for production. There will be some other big changes besides 60x60" tiles. Each tile will face a 30 arcmin circular "radius cut" like in MNRAS 2022. No more random downloads either. "Smart" tessellation plan will be made in form of JSON file, and the pipeline follows that plan. All stages and especially reductions will be separately reported per tile, per full dataset, and by each downstream step. No "main Parquets", no infrared catalogs, mainly lots of CSV files with lots of numbers. Southern Hemisphere will be excluded at tessellation plan phase. Probably all tiles that are not in PS1 coverage will be excluded, too. I might add the Earth's shadow statistics, but I'm bit hesitant: perhaps that topic has been sufficiently covered already.
I will continue running the new Vasco60 pipeline until I run out of disk space
