With the permission of Eric Feigelson, I am reproducing here this very useful post from the Astrostatistics and Astroinformatics Portal (ASAIP) blog on the many conferences on the discipline of Astrostatistics, and the growing number of powerful statistical software tools available to astronomers.
ASAIP readers may not realize the incredibly rich collection of scholarly conference, workshops and tutorials relating to astrostatisics and astroinformatics that are continuously being organized around the world. We try to keep the listing at http://asaip.psu.edu/meetings up to date, but any ASAIP member can add an entry. Let’s start with the meetings by the organizations affiliated with the ASAIP Web site:
- The International Astrostatistics Association affiliated with the International Statistical Institute (IAA/ISI) has an astrostatistics session within the giant World Statistics Congress in Rio de Janeiro at the end of July 2015. Astronomy was also featured at past WSCs in Dublin and Hong Kong.
- A week later in August, the International Astronomical Union Working Group on Astrostatistics and Astroinformatics (IAU/WGAA) is hosting a meeting Statistics and Exoplanets within the giant General Assembly in Honolulu HI. Exoplanets are a tremendously exciting new field of astronomy with many statistical challenges. The Honolulu meeting may be the largest gathering of astronomers in history.
- The American Astronomical Society Working Group in Astroinformatics and Astrostatistics (AAS/WGAA) has a session at its semiannual meeting in Seattle WA in January. It will discuss “Steps towards better curricula” to improve the methodological education of astronomers. The AAS has sessions on astrostatistics for many years.
- The American Statistical Association Interest Group in Astrostatistics (ASA/IGA) was formed at the Joint Statistical Meeting last summer in Boston. Astronomical sessions have been held at JSM meetings for years under the auspices of another Section, and are likely to continue into the future under the new IGA.
There are many other meetings around the world. There is a 1-day meeting with world-class speakers on statistical cosmology in London coming up in mid-December, a week-long `Cosmology on Safari’ meeting in South Africa in January, and presentations on statistical approaches Local Group galaxies in Ann Arbor MI in June. Big Data in astronomy is discussed by top methodologists at the National Optical Astronomical Observatory in Tucson AZ in March, and another meeting focuses on time domain issues in Santa Barbara CA in May.
And of course, there are many large well-established conferences on methodology in the next few months: IEEE conference on data mining in Shenzhen CN, ASE conference on Big Data in Cambridge MA, a SIAM conference on computational science, an AIENG conference on scientific computing, a SIAM conference on Data Mining in Vancouver CA, and so forth. The University of Toronto is hosting a unique 5-month program on Big Data. Links to all of these events are given on the ASAIP meetings page.
We have added a new folder to the Resources section of ASAIP: Astrostatistics and astroinformatics in the public eye. A special section of the December issue Significance Magazine, read by tens of thousands of statisticians worldwide, has 11 articles on astronomy and astrostatistics. The cover of the magazine portrays large meteor fragments impacting a large city and the accompanying article discusses the probability of this happening … in real life, not Hollywood movies! An article about the rise of astrostatistics, spanning `planet hunting to sky surveys’, appeared last month in Yahoo News, one of the most widely read Web sites on the Internet. A New York Times article several weeks ago interviewed a statistician, an astronomer and a psychologists on the value of Bayesian approaches to real-life problems. The Zooniverse citizen science project, mostly astronomy but now broadening to other fields, has involved >1 million people around the world. And most impressively, nearly a million people have watched astronomer Andy Connolly present a gripping TED talk on the revolution expected by the LSST telescope, which will produce an enormous movie of the variable sky.
Altogether, while we might view our interests as narrow scientific subfield, the public appears very receptive to the synergies underway between statistics, computation and astronomy.
With conferences, research and the public arena, there is no question that astrostatisticians and astroinformaticians are fluent with words and ideas. And individual researchers publish excellent scientific articles with sophisticated methods. But the commerce of advanced methodology from specialists to the wider research community has historically not been strong. We are pleased to say that practical methodological software is now emerging in several ways:
- After abortive attempts in the past, astronomy now has an effective repository for disciplinary computer codes, the Astronomy Source Code Library (ASCL). ASCL has ~1000 codes with a new one arriving daily. While many concern instrument-specific data analysis or methods for theoretical astrophysics, some are broadly applicable codes for advanced data analysis. Examples from submissions in the last few weeks include: `NDF: Extensible N-dimensional data format library’ for storage of N-dimensional arrays; `Anmap: Image and data analysis’ with modeling tools; ‘AMADA: Analysis of Multidimensional Astronomical DAtasets’ based on hierarchical clustering and PCA; `HOPE: Just-in-time Python compiler for astrophysical computation’ that translations Python code into C++ for numerical optimization; `MEPSA: Multiple Excess Peak Search Algorithm’ for analyzing multimodal behaviors in uniformly sampled time series; `DIAMONDS: high-DImensional And multi-MOdal NesteD Sampling’ for Bayesian inference with nested sampling Monte Carlo algorithm for multi-modal likelihoods.
- R, with >6000 user-provided add-on CRAN packages providing >100,000 statistical functions, is the largest public domain software environment for statistical analysis of data. R, still growing rapidly with >2M users, dominates fields such as genomics and econometrics but is rarely used in astronomy. However, its capabilities for astronomy are tremendous. R/CRAN, for example, has >110 packages for Bayesian computations and >80 for high performance computing. With only ~20 disciplinary packages for astronomy, it may appear that R is not ready for astronomical data analysis. But with a vast range of modern applied statistics already in R/CRAN, only a small range of disciplinary tools are needed for many analyses. CRAN packages include input/out of FITS files and a translation of much of the IDL Astronomy Users Library for time & coordinate transformation, precession and barycentric corrections, and other basic utilities.
- The Python language has many community-provided packages including for astronomy. In addition to considerable software infrastructure for astronomy in the astropy. This has a handful of statistical functions with a few dozen more in scipy.stats and scikits.statsmodels. Python has many strengths, but its coverage of statistics is far smaller than R’s. However, there is opportunity to import any of the >100,000 R/CRAN functions using the rpy2 interface (try “pip install rpy2”). Browse the interesting comparisons of Python and R by data scientists here.
We have had a close look at a small group of methodologically talented astronomers who are now developing software tools for the wider community. The COsmostatistics INitiative (COIN) within the International Astrostatistics Association (IAA) was formed in June 2014 at the IAU Symposium on statistical cosmology. COIN has since been very active under the leadership of Rafael de Souza of Eötvös Loránd University (Univ of Budapest). Two papers under the title The overlooked potential of generalized linear models in astronomy’ have been submitted for publication with associated R/CRAN and Python packages. Other projects on likelihood-free inference of cosmological parameters, Gaussian mixture models, unsupervised learning, and other topics are underway.
Even more recently, a new IAA group was formed to develop a software product called DEep DAta Look (DeDaLo). Under the leadership of Italian astronomers Roberto Casalegno, Claudia Travaglio, and Maurizio Busso, DeDaLo will be an interactive 3D environment to enhance the way scientific Big Data are displayed and therefore analyzed. It will be based on Blender, a professional 3D Open Source software product for animations and video games, using the Blender Game Engine (BGE) that provides a powerful high-level programming tool for interactive 3D user experience. DeDaLo will embed Python3 scientific tools in Blender to give powerful analysis tools in a video gaming 3D visualization environment.