SlideShare a Scribd company logo

Software Carpentry and the Hydrological Sciences @ AGU 2013

•
•
Aron Ahmadia
Aron Ahmadia

This document discusses bringing computational skills training to hydrologists through Software Carpentry workshops. It notes that while many hydrologists are focused on their research, computational methods are now essential. Software Carpentry teaches practical skills like the Unix shell, version control with Git, Python and R programming, and databases. These intensive, short workshops have been effective at training graduate students. The document encourages hydrologists to host their own workshops and support computational literacy by discussing code and practices in their papers.

1 of 38
Download to read offline
Software Carpentry and the Hydrological Sciences ! Aron Ahmadia Coastal and Hydraulics Lab US Army ERDC
Dark Matter, Public Health, and Scientific http://www.slideshare.net/gvwilson/dark-matter-publichealth-and-scientific-computing Computing This Talk is a Fork ! Greg Wilson
“Dark Matter Developers” Scott Hanselman (March 2012) ! [We] hypothesize that there is another kind of developer than the ones we meet all the time. We call them Dark Matter Developers. They don't read a lot of blogs, they never write blogs, they don't go to user groups, they don't tweet or facebook, and you don't often see them at large conferences... [They] aren't chasing the latest beta or pushing...limits, they are just producing. ! ! http://www.hanselman.com/blog/DarkMatterDevelopersTheUnseen99.aspx !3
We’re Not That Optimistic l 90% of hydrologists have their heads down − l l Doing hydrology instead of talking about using Charm++ to asynchronously distribute their CUDAaccelerated heterogeneous N-body Universe simulations Not because they don't want to do all that Star Trek stuff But because the majority of the field isn’t there yet. !4
Shiny Toys !5
Grimy Reality !6
So Here We Are Us Them !7
Surely You're Exaggerating 1. How many graduate students write shell scripts to analyze each new data set instead of running those analyses by hand? !8
Surely You're Exaggerating 2. How many of them use version control to keep track of their work and collaborate with colleagues? !9
Surely You're Exaggerating 3. How many routinely break large problems into pieces small enough to be - comprehensible, - testable, and - reusable? !10
Surely You're Exaggerating 3. How many routinely break large problems into pieces small enough to be - comprehensible, - testable, and - reusable? And how many know those are the same things? !11
We've Left the Majority Behind !12
We've Left the Majority Behind Other than Googling for things, the majority of scientists do not use computers more effectively today than they did 28 years ago. !13
“Not My Problem” Actually your biggest problem • • • Application Yesterday’s models are solving today’s million dollar problems, because nobody knows how to update them. Collaboration Hundreds of person-years of effort are redundantly combining to create a year’s worth of result Credibility Many published results can’t be trusted, because the scientific code isn’t modular, can’t be tested, and sometimes isn’t even released And Hydrology Loses !14
Jeff Hammond Your code will outlive your supercomputer. Most successful supercomputers lasts 5 years. The most successful codes may reach 5 decades. There are currently 40 year-old modeling codes still being used. Argonne Leadership Computing Facility !15
Where Are Your Goalposts? A hydrologist is computationally competent if she can build, use, validate, and share software to: •Manage software and data •Tell if it's working correctly •Find and fix problems when it isn’t •Keep track of what she’s done •Share work with others •Do all of these things efficiently !16
A Driver's License Exam l l Developed with the Software Sustainability Institute for the DiRAC Consortium Formative assessment − l Do you know what you need to know in order to get the most out of this gear? Pencils ready? Note: actual exam allows for several different programming languages, version control systems, etc. !17
A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Check out a working copy of the exam materials from a git repository. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !18
A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Use find and grep in a pipe to create a list of all .dat files in the working copy, redirect the output to a file, and add that file to the repository. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !19
A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Write a shell script that runs a legacy program for each parameter in a set. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !20
A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Edit the Makefile provided so that if any .dat file changes, analyze.py is re-run to create the corresponding .out file. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !21
A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Write four unit tests to exercise a function that calculates running sums. Explain why your four tests are most likely to uncover bugs in the function. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !22
A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Explain when and how a function could pass your tests, but still fail on real data. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !23
A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Do a code review of the legacy program from Q3 (about 50 lines long) and list the four most important improvements you would make. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !24
How Are We Doing? a) How well did you do? !25
How Are We Doing? a) How well did you do? b) How well do you think most hydrologists would do? !26
How Are We Doing? a) How well did you do? b) How well do you think most hydrologists would do? c) Do you think a hydrologist could use your software without having these skills? d) Or a grasp of the principles behind them? !27
What Has Worked l Target graduate students l Intensive short courses (2 days to 2 weeks) l Focus on practical skills !28
What Has Worked l Target graduate students l Intensive short courses (2 days to 2 weeks) l Focus on practical skills http://software-carpentry.org !29
!30
What We Teach l Unix shell how to automate repetitive tasks l Git and GitHub how to use distributed version control to manage, collaborate and share work l Python or R how to grow a program in a structured, modular, testable, reusable way l Databases the difference between structured and unstructured data !31
What We Teach l l l l l Unix shell Python Version control Testing Array computing, image processing, SQL, ... !32
“That's Merely Useful” l None of this is publishable any longer − Which means it's ineffective career-wise for academic computer scientists !33
“That's Merely Useful” l None of this is publishable any longer − l Which means it's ineffective career-wise for academic computer scientists But it works − Two independent assessments have validated the impact of what we do !34
Host a Workshop !35
Teach a Workshop l All our materials are CC-BY licensed l We will help train you Text ??? !36
Shine Some Light l l Include a few lines about your software stack and working practices in your scientific papers Ask about them when reviewing − − − Where's the repository containing the code? What's the coverage of your unit tests? How did you track provenance? !37
Links! Find this talk on Twitter: @ahmadia ! Me: http://aron.ahmadia.net ! Software Carpentry: http://software-carpentry.org ! US Army Engineer Research and Development Center: http://www.erdc.usace.army.mil/ !38

More Related Content

Software Carpentry and the Hydrological Sciences @ AGU 2013

  • 1. Software Carpentry and the Hydrological Sciences ! Aron Ahmadia Coastal and Hydraulics Lab US Army ERDC
  • 3. “Dark Matter Developers” Scott Hanselman (March 2012) ! [We] hypothesize that there is another kind of developer than the ones we meet all the time. We call them Dark Matter Developers. They don't read a lot of blogs, they never write blogs, they don't go to user groups, they don't tweet or facebook, and you don't often see them at large conferences... [They] aren't chasing the latest beta or pushing...limits, they are just producing. ! ! http://www.hanselman.com/blog/DarkMatterDevelopersTheUnseen99.aspx !3
  • 4. We’re Not That Optimistic l 90% of hydrologists have their heads down − l l Doing hydrology instead of talking about using Charm++ to asynchronously distribute their CUDAaccelerated heterogeneous N-body Universe simulations Not because they don't want to do all that Star Trek stuff But because the majority of the field isn’t there yet. !4
  • 7. So Here We Are Us Them !7
  • 8. Surely You're Exaggerating 1. How many graduate students write shell scripts to analyze each new data set instead of running those analyses by hand? !8
  • 9. Surely You're Exaggerating 2. How many of them use version control to keep track of their work and collaborate with colleagues? !9
  • 10. Surely You're Exaggerating 3. How many routinely break large problems into pieces small enough to be - comprehensible, - testable, and - reusable? !10
  • 11. Surely You're Exaggerating 3. How many routinely break large problems into pieces small enough to be - comprehensible, - testable, and - reusable? And how many know those are the same things? !11
  • 12. We've Left the Majority Behind !12
  • 13. We've Left the Majority Behind Other than Googling for things, the majority of scientists do not use computers more effectively today than they did 28 years ago. !13
  • 14. “Not My Problem” Actually your biggest problem • • • Application Yesterday’s models are solving today’s million dollar problems, because nobody knows how to update them. Collaboration Hundreds of person-years of effort are redundantly combining to create a year’s worth of result Credibility Many published results can’t be trusted, because the scientific code isn’t modular, can’t be tested, and sometimes isn’t even released And Hydrology Loses !14
  • 15. Jeff Hammond Your code will outlive your supercomputer. Most successful supercomputers lasts 5 years. The most successful codes may reach 5 decades. There are currently 40 year-old modeling codes still being used. Argonne Leadership Computing Facility !15
  • 16. Where Are Your Goalposts? A hydrologist is computationally competent if she can build, use, validate, and share software to: •Manage software and data •Tell if it's working correctly •Find and fix problems when it isn’t •Keep track of what she’s done •Share work with others •Do all of these things efficiently !16
  • 17. A Driver's License Exam l l Developed with the Software Sustainability Institute for the DiRAC Consortium Formative assessment − l Do you know what you need to know in order to get the most out of this gear? Pencils ready? Note: actual exam allows for several different programming languages, version control systems, etc. !17
  • 18. A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Check out a working copy of the exam materials from a git repository. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !18
  • 19. A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Use find and grep in a pipe to create a list of all .dat files in the working copy, redirect the output to a file, and add that file to the repository. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !19
  • 20. A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Write a shell script that runs a legacy program for each parameter in a set. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !20
  • 21. A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Edit the Makefile provided so that if any .dat file changes, analyze.py is re-run to create the corresponding .out file. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !21
  • 22. A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Write four unit tests to exercise a function that calculates running sums. Explain why your four tests are most likely to uncover bugs in the function. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !22
  • 23. A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Explain when and how a function could pass your tests, but still fail on real data. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !23
  • 24. A Driver's License Exam 1. 2. 3. 4. 5. 6. 7. Do a code review of the legacy program from Q3 (about 50 lines long) and list the four most important improvements you would make. Could do it easily Could struggle through 1 0.5 Wouldn’t know where to start 0 Don’t understand the question -1 !24
  • 25. How Are We Doing? a) How well did you do? !25
  • 26. How Are We Doing? a) How well did you do? b) How well do you think most hydrologists would do? !26
  • 27. How Are We Doing? a) How well did you do? b) How well do you think most hydrologists would do? c) Do you think a hydrologist could use your software without having these skills? d) Or a grasp of the principles behind them? !27
  • 28. What Has Worked l Target graduate students l Intensive short courses (2 days to 2 weeks) l Focus on practical skills !28
  • 29. What Has Worked l Target graduate students l Intensive short courses (2 days to 2 weeks) l Focus on practical skills http://software-carpentry.org !29
  • 30. !30
  • 31. What We Teach l Unix shell how to automate repetitive tasks l Git and GitHub how to use distributed version control to manage, collaborate and share work l Python or R how to grow a program in a structured, modular, testable, reusable way l Databases the difference between structured and unstructured data !31
  • 32. What We Teach l l l l l Unix shell Python Version control Testing Array computing, image processing, SQL, ... !32
  • 33. “That's Merely Useful” l None of this is publishable any longer − Which means it's ineffective career-wise for academic computer scientists !33
  • 34. “That's Merely Useful” l None of this is publishable any longer − l Which means it's ineffective career-wise for academic computer scientists But it works − Two independent assessments have validated the impact of what we do !34
  • 36. Teach a Workshop l All our materials are CC-BY licensed l We will help train you Text ??? !36
  • 37. Shine Some Light l l Include a few lines about your software stack and working practices in your scientific papers Ask about them when reviewing − − − Where's the repository containing the code? What's the coverage of your unit tests? How did you track provenance? !37
  • 38. Links! Find this talk on Twitter: @ahmadia ! Me: http://aron.ahmadia.net ! Software Carpentry: http://software-carpentry.org ! US Army Engineer Research and Development Center: http://www.erdc.usace.army.mil/ !38