Project

General

Profile

SG separation challenge » History » Version 19

William Wester, 12/18/2013 04:48 PM

1 1 Ignacio Sevilla
h1. SG separation challenge
2 1 Ignacio Sevilla
3 1 Ignacio Sevilla
Now that several people are testing their own approaches:
4 1 Ignacio Sevilla
5 19 William Wester
* Cut-based with DESDM info (Eli, Diego, Nacho, Ryan, William...).
6 1 Ignacio Sevilla
* Multi-class (Maayane)
7 1 Ignacio Sevilla
* Random Forests (Ryan)
8 1 Ignacio Sevilla
* Boosted Decision Trees (Nacho)
9 1 Ignacio Sevilla
* Alternative Neural Network with probabilistic output (Chris Bonnett).
10 13 Basilio Santiago
* Probability based on spread model and photometry (DES-Brazil)
11 14 Basilio Santiago
* Others...
12 1 Ignacio Sevilla
13 1 Ignacio Sevilla
I think the time is right and the codes are mature to launch a specific SG separation challenge, mimicking the successful photo-z WG exercise.
14 1 Ignacio Sevilla
15 1 Ignacio Sevilla
We have to establish:
16 1 Ignacio Sevilla
17 1 Ignacio Sevilla
* The training/validation/testing sample (COSMOS, others).
18 17 Ignacio Sevilla
I have prepared a 70/30 training/testing with the deep COSMOS field matched to ACS imaging. About 280 parameters, up to each tester to choose which.
19 17 Ignacio Sevilla
Besides new datasets, also consider shallower COSMOS. Also consider fixed set of parameters as Eduardo suggests. Also need to add SLR corrections though I think not very important now.
20 6 Ignacio Sevilla
* Only stars and galaxies? What about QSOs, image artifacts?
21 17 Ignacio Sevilla
Star/galaxy for round 1.
22 3 Ignacio Sevilla
* The metrics (Fixed cut, Fixed purity, Fixed Efficiency, ROC -- see example below).
23 17 Ignacio Sevilla
I would prefer to use ROC, i.e., completeness vs purity curve formed changing the threshold.
24 7 Eli Rykoff
* SVA1 systematics: correlations with depth, Galactic latitude, seeing, etc.
25 4 Ignacio Sevilla
* Who/how to run it.
26 17 Ignacio Sevilla
I suggest each group providing an output file with id (or ra,dec on first round) plus galaxy probability or binary value.
27 8 Ignacio Sevilla
* Is there any gain combining them (a committee)?
28 1 Ignacio Sevilla
* The schedule.
29 9 Maayane Soumagnac
30 18 Ignacio Sevilla
[[des-sci-verification:SG_separation_challenge_details|Test Details]]
31 18 Ignacio Sevilla
32 9 Maayane Soumagnac
h1. The metric
33 9 Maayane Soumagnac
34 9 Maayane Soumagnac
We suggest to use the same metric as in the DES star/galaxy separation (on simulation) paper (arXiv:1306.5236).
35 9 Maayane Soumagnac
36 10 Maayane Soumagnac
h2. Completeness and Purity provided by a given classifier
37 10 Maayane Soumagnac
38 9 Maayane Soumagnac
We define the parameters used to quantify the quality of a star/galaxy classifier. For a given class of objects, X (stars or galaxies), we distinguish the surface density of well classified ob jects, N_X , and the misclassified objects, M_X .
39 9 Maayane Soumagnac
40 9 Maayane Soumagnac
* The galaxy completeness c^g is defined as the ratio of the number of true galaxies classified as galaxies to the total number of true galaxies. 
41 9 Maayane Soumagnac
* The stellar contamination f_s is defined as the ratio of stars classified as galaxies to the total amount of ob jects classified as galaxies. 
42 9 Maayane Soumagnac
* The purity p^g is defined as 1-f_s
43 9 Maayane Soumagnac
44 11 Maayane Soumagnac
!{width:400px}metric.png! 
45 10 Maayane Soumagnac
46 10 Maayane Soumagnac
h2. Plots
47 10 Maayane Soumagnac
48 10 Maayane Soumagnac
Bellow are three different plots we suggest to use to assess the performances of each classifier.
49 10 Maayane Soumagnac
50 10 Maayane Soumagnac
h3. Histograms
51 10 Maayane Soumagnac
52 10 Maayane Soumagnac
Example, on simulations, from arXiv:1306.5236
53 11 Maayane Soumagnac
!{width:400px}histo.png!
54 10 Maayane Soumagnac
55 10 Maayane Soumagnac
h3. purity as a function of magnitude (for fixed completeness, the threshold/cut is let free)
56 1 Ignacio Sevilla
57 12 Eli Rykoff
!{width:800px}Emma.png!
58 10 Maayane Soumagnac
59 16 Ignacio Sevilla
!{width:400px}sg_separation_purity_vs_magauto_50.0_efficiency.png! !{width:400px}sg_separation_purity_vs_magauto_90.0_efficiency.png! 
60 16 Ignacio Sevilla
61 15 Ignacio Sevilla
h3. completeness as a function of magnitude (for fixed purity, the threshold/cut is let free )
62 15 Ignacio Sevilla
63 15 Ignacio Sevilla
!{width:400px}sg_separation_efficiency_vs_magauto_95.0_purity.png! !{width:400px}sg_separation_efficiency_vs_magauto_99.0_purity.png!