Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
The search for single transits
Search
Dan Foreman-Mackey
May 08, 2015
Science
1
300
The search for single transits
My short talk from the Sagan Fellows Symposium at Caltech
Dan Foreman-Mackey
May 08, 2015
Tweet
Share
More Decks by Dan Foreman-Mackey
See All by Dan Foreman-Mackey
Open software for Astronomical Data Analysis
dfm
0
110
Open Software for Astrophysics, AAS241
dfm
2
440
My research talk for CCA promotion
dfm
1
740
Astronomical software
dfm
1
690
emcee-odi
dfm
1
590
Exoplanet population inference: a tutorial
dfm
3
420
Data-driven discovery in the astronomical time domain
dfm
6
680
TensorFlow for astronomers
dfm
6
710
How to find a transiting exoplanets
dfm
1
440
Other Decks in Science
See All in Science
機械学習による確率推定とカリブレーション/probabilistic-calibration-on-classification-model
ktgrstsh
2
160
Non-Gaussian methods for causal discovery
sshimizu2006
0
260
Reaping the Benefits of Ritual and Routine
arthurdoler
PRO
0
160
山形とさくらんぼに関するレクチャー(YG-900)
07jp27
1
190
Introduction to Graph Neural Networks
joisino
PRO
4
1.9k
Celebrate UTIG: Staff and Student Awards 2024
utig
0
260
20分で分かる Human-in-the-Loop 機械学習におけるアノテーションとヒューマンコンピューターインタラクションの真髄
hurutoriya
4
2k
【健康&筋肉と生産性向上の関連性】 【Google Cloudを企業で運用する際の知識】 をお届け
yasumuusan
0
290
Machine Learning for Materials (Lecture 1)
aronwalsh
1
1.9k
マテリアルズ・インフォマティクスの先端で起きていること / What's Happening at the Cutting Edge of Materials Informatics
snhryt
1
110
Boil Order
uni_of_nomi
0
100
ウェーブレットおきもち講座
aikiriao
1
770
Featured
See All Featured
Speed Design
sergeychernyshev
22
430
Teambox: Starting and Learning
jrom
131
8.7k
VelocityConf: Rendering Performance Case Studies
addyosmani
322
23k
Statistics for Hackers
jakevdp
794
220k
Principles of Awesome APIs and How to Build Them.
keavy
125
16k
Optimizing for Happiness
mojombo
375
69k
Scaling GitHub
holman
458
140k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
29
2.6k
A Tale of Four Properties
chriscoyier
155
22k
Building Your Own Lightsaber
phodgson
101
6k
Happy Clients
brianwarren
96
6.6k
JavaScript: Past, Present, and Future - NDC Porto 2020
reverentgeek
45
4.8k
Transcript
Single the search for Transits Dan Foreman-Mackey NYU→UW // github.com/dfm
// @exoplaneteer // dfm.io
David W. Hogg NYU Bernhard Schölkopf MPI-IS
Population Inference
treatment of false positives, dependent parameters, uncertainties & selection effects
open source tools applicable to all existing & future exoplanet missions occurrence rate period, radius, mass, eccentricity, multiplicity, mutual inclination, etc. Flexible & robust inference of the exoplanet population
1 catalog of planet (candidates) measurement of completeness 2 3
measurement of precision Ingredients of a population inference
101 102 orbital period [days] 100 101 planet radius [R
] Data from NASA Exoplanet Archive
101 102 orbital period [days] 100 101 planet radius [R
] Data from NASA Exoplanet Archive
100 101 102 103 104 105 orbital period [days] 100
101 planet radius [R ] Data from NASA Exoplanet Archive
10 100 f 10 30 100 N detection S/N threshold
# of detectable single transits Extrapolated from Dong & Zhu (2013)
How to find a Transiting Planet the traditional way…
1 de-trending grid search in period, phase, and duration 2
3 vetting of candidates How to find a (periodic) transit signal
False Alarms & False Positives
How to find a Transiting Planet the Planet Hunters way…
None
Can we Teach the Machine to Learn™?
Bernhard Schölkopf MPI-IS Get rid of the pipeline!
no_transit transit vs. 1 0 1 time [days] 1 0
1 time [days] Supervised Classification
Supervised Classification
Random Forest™ Classification NYC LA 10 8 NYC LA 7
2 NYC LA 3 6 Raining Sunny Car Subway NYC LA 0 6 NYC LA 3 0 NYC LA 0 2 NYC LA 7 0 Beach Park decision tree
Random Forest™ Classification NYC LA 10 8 NYC LA 7
2 NYC LA 3 6 Raining Sunny Car Subway NYC LA 0 6 NYC LA 3 0 NYC LA 0 2 NYC LA 7 0 Beach Park decision tree
light curve sections simulated transits held-out light curve features training
set test set
200 400 600 800 1000 1200 1400 time [KBJD] 0.003
0.002 0.001 0.000 0.001 0.002 0.003 0.004
no_transit transit vs. 1 0 1 time [days] 1 0
1 time [days]
scikit-learn.org
Preliminary Results
light curves false positives transit candidate 3,000 273 1
9821962 9847647 10544712 9834736 9763612 9763027 2 0 2 10554152
2 0 2 9776926 time since transit [days] 9821962 9847647 10544712 9834736 9763612 9763027 2 0 2 10554152 2 0 2 9776926 time since transit [days] 10602068 10286702 10518652 9775416 9821962 9847647 10544712 9834736 9763612 9763027 False Positives
3.0 3.3 3.6 3.9 log10 P/day 0.21 0.22 0.23 0.24
t0 830.8 KBJD [hr] 0.58 0.60 0.62 b 1.2 1.8 2.4 3.0 Rp [RJ ] 0.15 0.30 0.45 0.60 e 3.0 3.3 3.6 3.9 log10 P/day 0.21 0.22 0.23 0.24 t0 830.8 KBJD [hr] 0.58 0.60 0.62 b 0.15 0.30 0.45 0.60 e 824 826 828 830 832 834 836 838 0.90 0.92 0.94 0.96 0.98 1.00 1.02 824 826 828 830 832 834 836 838 0.90 0.92 0.94 0.96 0.98 1.00 1.02 824 826 828 830 832 834 836 0.90 0.92 0.94 0.96 0.98 1.00 1.02
No good model of the non-transits…
Temporary solution: Template likelihoods
1 can discover single transits using supervised classification false positives
are still a problem (but maybe less) 2 3 would like to combine method with realistic noise model Conclusions