Why your field needs a hack week

The skills that broadly fall under the umbrella of data science are becoming increasingly important to modern-day science. As demands upon researchers to apply and implement new methodologies and data analysis techniques rise, training scientists in these methods is becoming more of a priority.

The week of this talk, BIDS is hosting Astro Hack Week, a five-day mix of summer school, unconference, and hackathon. It is one approach to improving data literacy in astronomy while at the same time providing a collaborative venue for researchers to share ideas and their approaches to problems, work on research projects, and learn new concepts that go beyond those generally taught in astronomy. In this talk, I will give an overview of the ideas that underlie Astro Hack Week, highlight some results from this week’s workshop, and try to convince you that your field needs a hack week too.

Transcript

1. Why your field needs a _____ hack week Daniela Huppenkothen NYU Center for Data Science NYU Center for Cosmology and Particle Physics ! Tiana_Athriel " dhuppenkothen astro bio politics chemistry social geo physics
2. This week: #AstroHackWeek @ BIDS
3. • 5 days • 50 researchers • all of astronomy represented • tutorials, lectures, hacking • 28 projects in progress #AstroHackWeek @ BIDS
4. Why?
5. Data is exploding 700 600 500 400 IRSA holdings (TB) 300 200 100 2008 0 2009 IRSA general Spitzer WISE 2010 2011 2012 2013 2014 Growth in the Scientific Data Holdings of IRSA, Projected to 2014 Chart courtesy of IRSA
6. Credit: CERN/LHC/CMS www.followthehashtag.com Credit: SDSS Credit: William Putman, NASA/Goddard Data is exploding
8. Our data is getting more complex 960,000 terabytes per day!
9. • heterogeneous • complex • dirty • high volume • high velocity Modern data is
10. by Calvin.Andrus
11. academia is lagging behind industry in data-driven research • http://jakevdp.github.io/blog/2013/10/26/big-data-brain-drain/ • https://jakevdp.github.io/blog/2014/08/22/hacking-academia/
12. How do we encourage researchers to learn and adopt new methods?
13. Traditional conferences are segregated by sub discipline
14. How do we get researchers to talk to each other about methods?
15. • teach new data science methods • increase connectedness • foster exchange between disciplines • build networks • promote open science Objectives
17. What AstroHackWeek is not
18. a conference
20. “data-driven communication” vs “science-driven communication”
22. a summer school
23. summer schools are “top down”
24. everyone learns from everyone hack week principle:
25. How does it work?
26. Ingredients: • tutorials • break-out sessions • hacking
27. Tutorials
28. AstroHackWeek tutorials are scarier for lecturers than participants! solution: peer learning
29. Tutorials
30. Break-out Sessions
32. Hacking
33. “hack” A hack is a small project with a very clear goal, which should be completed by the end of the time initially allocated to it
34. “hack” To cobble together a quick and dirty solution to a problem
35. “hack” To simplify (or take a quick approach to solving) a complex problem
36. Hacking
37. Hacking https://hackpad.com/Astro-Hack-Week-2016- Hack-Index-PlmbAyVqqtQ
38. Hacking
39. Hacking
40. Encouraging Open Science
41. Diversity is excellence
42. Diversity beginners — experts: allows learning
43. Diversity undergrads — senior faculty: build networks
44. Diversity academic sub disciplines: foster exchange
45. Diversity diversity in background: new perspectives + ideas
46. AstroHackWeek is only as successful as its participants
47. We want a diverse set of participants from different backgrounds and all academic subdisciplines of astronomy, and also a wide range of technical backgrounds from beginners to experts in any of the topics we are interested in teaching and working on, and also a wide range of academic seniority
48. Goals: • optimise mix of participants • reduce bias • transparency • accountability
49. ???
50. “You’re a data science fellow. Do data science with this!” — Brian McFee
51. Computers are better at complex optimisation than humans
52. Step 1: Select all appropriate applicants
53. Step 2a: pre-select (a few) experienced hackers
54. Step 2b: use algorithm to break ties conditioned on the ideal mix of participants
55. Statistics Experience
56. Demographics
57. Entrofy https://github.com/dhuppenkothen/entrofy Daniela Huppenkothen, Brian McFee, Taisya Koptova, Katalin Takats, Kelle Cruz (developed at AstroHackWeek 2015)
58. Does AstroHackWeek work? Yes, but …
59. “What did you learn at Astro Hack Week?” “Machine learning (and where the resources are if I want to learn more)“ “Not to be too afraid of Bayesian methods“ “The evening "debate" about corrected an important misunderstanding I had about model selection”
60. “What did you learn at Astro Hack Week?” “[…] learned about new tools to use, new methods to try, and a lot of good practices” “team coding” “I was really able to improve my programming skills and habits after this workshop” “I learned a lot about profiling and commenting code, and methods for improving performance”
61. “What did you learn at Astro Hack Week?” “The main thing I took from AstroHackWeek is the fact that I'm not alone […]” “Working with Brewer on transdimensional sampling was extremely helpful for our current work.”
62. “I believe that AstroHackWeek was useful for my future career”
63. “I believe that I learned skills at AHW that are applicable outside academia”
64. “I feel like I learned things which improve my day to day research”
65. “I feel like AHW has made me more comfortable with doing open science”
66. Projects
67. Uncovering planets and stellar activity using radial velocities 88, A31 (2016) er- es: e , We of by lly ve el- an he es nd re- en are .g. ms, ill 6). o- Np planets N data points µP vkep i Np µK e K P ⌘4 ⌘1 ⌘2 ⌘3 vsys ⌃ ! s ti wP vi i Fig. 1. Representation of the relations between parameters and obser- vations in our RV model, as a probabilistic graphical model. An arrow between two nodes indicates the direction of conditional dependence. The circled nodes are the parameters of the model, whose joint dis- tribution is sampled by the Markov chain Monte Carlo (MCMC) al- gorithm. The double circled node v i represents the observed RVs. The filled nodes represent deterministic variables: if these variables have parent nodes (vkep i and ⌃), they are given by a deterministic function of Faria, Haywood, Brewer et al, 2016 A&A 588, A31 0 1 2 3 4 5 6 7 8 9 10 Number of planets 0 500 1000 1500 2000 2500 3000 Number of posterior samples p(2) p(1) ¥ Œ p(3) p(2) ¥ 4.49 Fig. 2. Posterior distribution for the number of planets Np . The counts are number of posterior samples in models with a given number of planets. The two ratios of probabilities between models with 1, 2, and 3 planets are highlighted; note that p (0) = p (1) = 0. (se the pa am wh ur wh rit Pb ce aro str pe (se an
68. Diffraction Microscopy David Hogg, Dalya Baron, Matt Mechtley
69. MLCS2k2 sncosmo integration Kyle Barbary, Saurabh Jha
70. Cosmic Web Classification with Machine Learning Thales Gutcke, Elise Jennings
71. New Tutorials Phil Marshall, Kyle Barbary
72. UrbanGoggles https://hackpad.com/ UrbanGoggles-7kSFBc1yuec
73. Collaboratr Mike Baumer, Usman Khan, Casey Law, Ben Nelson
74. HoggBot Taisya Koptova, Boris Leistedt, Dan Foreman-Mackey
75. Lessons Learned
76. Participants come to AstroHackWeek with vastly different expectations and goals Lessons Learned
77. Student travel support is crucial Lessons Learned
78. Space matters Lessons Learned
79. Lessons Learned The Impostor Syndrome is real and hangs out at AstroHackWeek, too!
80. Lessons Learned all work and no food/coffee makes hackers cranky
81. Lessons Learned Evenings are important (hacking continues)
82. Things we don’t know • How can we measure success? • What is the ideal set of participants for Astro Hack Week? • What is the right balance between learning and project work? • How can we effectively mitigate impostor syndrome?
83. Should you run a hack week? we have more data than we know what to do with our data is complex we need better/new methods collaboration and networking is important to us we want to make open science a priority
84. Should you run a hack week? Yes
87. Resources • http://www.astrohackweek.org • https://github.com/AstroHackWeek • https://hackpad.com/Astro-Hack-Week-2016-Hack-Index- PlmbAyVqqtQ • https://danielahuppenkothen.wordpress.com • https://neurohackweek.github.io • https://geohackweek.github.io • http://github.com/dhuppenkothen/entrofy • https://medium.com/@dalcashdvinsky/the-horror-of-hack- days-52c6b52cfc3b#.biklgcpd3 • Come and talk to us!
88. (special thanks to: Kyle Barbary, Phil Marshall, David W. Hogg, Jake Vanderplas, Brian McFee) Questions?