460 likes | 575 Views
You think you are a young star?. New Measures for Different Transients. Ashish Mahabal, Caltech aam at astro.caltech.edu 4th Gaia Science Alerts Workshop IAP, Paris, 19-21 June 2013. Collaborators. Gaia collaboration Lukasz Wyrzykowski Sergey Koposov , Gerry Gilmore, Simon Hodgkin
E N D
You think you are a young star? New Measures for Different Transients Ashish Mahabal, Caltech aam at astro.caltech.edu 4th Gaia Science Alerts Workshop IAP, Paris, 19-21 June 2013
Collaborators Gaia collaboration Lukasz Wyrzykowski Sergey Koposov, Gerry Gilmore, Simon Hodgkin SAMSI: Julian Faraway, Jiayang Sun, JogeshBabu, Lingsong Zhang, Grace Wang, Xiaofeng Wang • Caltech • Alex Ball • George Djorgovski • CiroDonalek • Andrew Drake • Matthew Graham • KartikSaxena • Roy Williams • JPL • Thomas Fuchs • Mike Turmon India: AjitKembhavi, Sajeeth Philip South Africa: SudhanshuBarway • Plus at various other institutes all over, but especially in US, India and Italy LSST Ashish Mahabal
Overview • Objectives • Classification, in Real-time, using minimal data • Applicable to archives and follow-up prioritization • Challenges • Heterogeneity of data sources • CRTS (today ~10), LSST(soon ~10^6), minimal overlap (e.g. DASCH), part of larger set of parameters • Large and massive amount of light curves • Missing data, measurement errors and irregularly sampled data, …. • Data Sets • Lightcurves+, necessary bits from archives, images
Data • Transients from CRTS • Mostly non-variables: Objects @ random locations • Brighter samples of CVs and RR-Lyrae – important for connecting datasets (e.g. many brighter CRTS objects will saturate LSST, just like almost all DASCH sources are saturated in CRTS) Soon: Gaia, and LSST simulations
Data Characteristics Classifying (all) transients (in real time) is hard • Too many ‘ordinary’ transients • Finding needles in a hay stack • Too many possible ‘parameters’ • e.g. colors, positions, flux CRTS --> LSST
Digital snapshots -> digital movies of the sky CRTS: PIs George/Andrew 500 M lightcurves (time series) available for analysis. Available soon also from IUCAA, India
Challenge 1: Characterize/Classify as much with as little data as possible We concentrate here on lightcurves (time series)
Challenge 2: Only a small fraction are rare* CRTS statistics as of Jun 2013: http://nesssi.cacr.caltech.edu/catalina/Stats.html • Current Status: • About 1 strong (but mostly ‘ordinary’) transient/106 sources by machine • High threshold to pick most dramatic transients (identification by humans) • Future: • With LSST, a million transients will be found per night, a good reason why we need automatic classification algorithms Ast/Flr SNe
Challenge 3: A Variety of Parameters Not all parameters are always present leading to swiss-cheese like data sets • Discovery: magnitudes, delta-magnitudes • Contextual: • Distance to nearest star • Magnitude of the star • Color of that star • Normalized distance to nearest galaxy • Distance to nearest radio source • Flux of nearest radio source • Galactic latitude • Follow-up • Colors (g-r, r-I, i-z etc.) • Prior classifications (event type) • Characteristics from light-curve • Amplitude • Median buffer range percentage • Standard deviation • Stetson k • Flux percentile ratio mid80 • Prior outburst statistic http://ki-media.blogspot.com/ • New lightcurve-based parameters: • Whole curve measures • Fitted curve measures • Residual from fit measures • Cluster measures • Other
Our Approaches • Methods (recall our objective: Classification): • 1. Modern EDA before classification on stats, lightcurvesin 1-d and high-d (graphical computation) • Improvement from 4 directions: • 1. Better with new derived statistics • 2. Better classification procedure (single, ensemble) • 3. Better with previously ignored information • ‘semi-supervised’ learning • 4. Better in terms of using less or incremental approach • Notes: Classification based on derived statistics or entire curve (2-4) • 3. Methodology Development
EDA on a sub-group: active galactic nuclei, which includes blazar
SED of boxplots for CRTS flares CRTS, SDSS, 2MASS, WISE mags
Derive new statistics • How? • Fit curves (by FDA, NP, Gaussian process modeling) • Functional Data Analysis: registration • Non-Parametric (Regression): incorporate known variances • GPM: use the known variances to build the prior • Residuals: • Variability, outliers/signals, … • Others
Generation of new summary measures Modeled Curve Fitted Summary measures Residuals Summary measures Clusters of observations In 30 minute groups of 4 Summary measures
(Old/)New Summary Statistics • Whole curve measures Median magnitude (mag); mean of absolute differences of successive observed magnitude; the maximum difference magnitudes • Fitted curve measures Scaled total variation scaled by number of days of observation; range of fitted curve; maximum derivative in the fitted curve • Residual from fit measures The maximum studentized residual; SD of residuals; skewness of residuals; Shapiro-Wilk statistic of residuals • Cluster measures Fit the means within the groups (up to 4 measurements); and then take the logged SD of the residuals from this fit; the max absolute residuals from this fit; total variation of curve based on group means scaled by range of observation • Other
Relative significance of parameters Linear trend: sign(linear trend) × log(linear trend| + 1e−06) sign(linear trend) ×√{|linear trend|} med_buf_range_per: −log(1 − med_buf_range_per) Kurtosis: log(3 + kurtosis) Parameters from Richards et al.
Available Data with non-variables and 7 transient types Random split Training Set N=2480 Test Set N=1240 Percent correctly classified in the test set: Others: Multinomial logist DA + New Ensembles
Whole Curve Comparisons • PfClust -> PfClassification • Functional Centroid Method(FCC) • Model m(x) and of the whole curves for each class • Develop Simultaneous Confidence Bands for each m(x) • Define a functional distance measure between curves • Classify a new curve to one of the existing classes or a new class of curves based on the distances
Development of Functional Method Exploration Step: are they different and separable? • Directly estimate the (pair-wise) mean difference between classes • Bootstrap method to estimate the (point-wise) confidence intervals.
Using domain knowledge effectively As part of the general theme of classifying transients and variables, it is important to separate SNe from non-SNe Simple parameters that can be used: • Distance to nearest star (and perhaps color) • Galaxy proximity (normalized) • Archival lightcurve (including with upper limits) Ashish Mahabal
Proximity to a galaxy is useful in marking SNe • Often there is confusion when more than one galaxy is present nearby • As a result distance has to be normalized • Which radius to consider is also an important consideration • Misclassification of S/G possible, so look at nearest stars • Coincident stars will be a direct NO • Dwarf galaxies (low SNR – not catalogued – have to be considered) Only a small fraction have radio-counterparts – so that too is a NO (but if yes, its likely to be very interesting) Ashish Mahabal
Definite SN – but in which galaxy? The transient (not seen) is at the center) Ashish Mahabal
Separation of SNe and non-SNe normalized Based on peaks 80-90% completeness Ashish Mahabal
R CorBor – has two distinct states Carbon rich with dust formation episodes Sumin Tang
Period Changing RRL A Drake Neither Blazhko, nor RRd.
Period changing RR Lyrae A Drake Split data in two parts. Two separate, discrete periods separated by 100 day gap. RRab -> RRc?
Characterizing measures - I • Variability: • Abbe, von Neumann, Stetson J, K, L, reduced chi-square, Kendall • Morphology: • Wozniak (2000) consecutive statistic • Cumulative sum range • Moments: mean, variance, skew, kurtosis • Median absolute deviation • Thiel-Sen estimator of median slope • Periodicity: • Period by conditional entropy • Optimal Fourier Decomposition using CE (F-test) • Weighted wavelet z-transform around CE period • Autocorrelation: • Kendall τ statistic • Durbin-Watson statistic • Detrended fluctuation analysis (long memory or 1/f processes) • Hurst exponent (long term memory) • ZCDF: ACF(τ0) = 0, KACF • Granger Causality Analysis (temporal dependency) M Graham
Characterizing measures - II • Processes: • Teraesvarta neural network test (nonlinearity) • Lyapunov exponent (chaos) • Slepian wavelet (characteristic timescale) • HMM: • Inhomogeneous mixture (AIC/BIC) • Continuous (AIC/BIC) • CAR(0) / CAR(p) (AIC) M Graham
Summary • Developing new derived statistics • Early Characterizing needed for selecting rare ones • Allowing for incremental classification • Characterizing based on domain knowledge • Public datasets like CRTS/skydot important Gaia follow-up and LSST simulations to be linked soon