Copy This Lecture!

Software Carpentry Overview by Software Carpentry is licensed under a Creative Commons Attribution 3.0 Unported License.

History

• Founded by Greg Wilson in 1998, teaching scientists how to use supercomputers at LANL.
• Open sourced materials 2004-present
• Currently funded by the Sloan Foundation and the Mozilla Foundation

What We Teach

• Unix Command Line Interface (Shell)
• Shell automation
• Version Control
• Python
• Testing

What We Actually Teach

• A program is just another piece of lab equipment
• Programming is a human activity
• Little pieces loosely joined
• Let the computer repeat it
• Paranoia makes us productive
• Better algorithms beat better hardware

How to THINK like a programmer

Who We Teach

Who We Are

• Adina Howe (Argonne / Michigan State U)
• Will Trimble (Argonne)
• Kirsten Hofmockel (ISU)
• volunteers

Our Goals for You

We will take you on a tour of:

• Managing and sharing Software, Data, and Manuscripts with Git
• Automating things with the shell
• Practical Programming with Python
• Scientific Computing with Python (numpy, matplotlib)

Be fluent in multiple languages

You speak multiple languages when interacting with a computer. Choosing to use a new tool, library, or language can be similar to learning a new language:

• There is a high initial startup cost as you learn vocabulary, grammar, and idioms sum(x*y for x,y in itertools.izip(x_vector, y_vector))
• But once you have gained some fluency, you will find yourself capable of new things!
• You will learn faster by observing and working with others who are more skilled than you
• Aim for languages and tools that allow you to express your models and manage your data simply.

Make it work right first, make it fast later.

• "Premature optimization is the root of all evil." -- Donald Knuth
• Directing your attention to making it use less disk / less memory / less time from the start is wrongly directed attention.

Increase debugging bandwidth

• REPL (read-eval-print-loop) environments tighten the coupling between the code you write and the results you see, increasing productivity.
• Development environments and debuggers give you more information at once
• Test your procedures on subsets of your data so that you learn whether it works faster

Automate common actions by saving simple blocks of code into scripts

• A script is a set of commands organized into a single file
• The script is the basest unit of scientific programming, you should be comfortable writing these whenever you want to save or otherwise document or repeat your actions
• Use scripts to explore new ideas, they are easy to write and throw away
• Don't repeat commands into your REPL, save them to a script

Refactor commonly used blocks of code into functions

• Eventually, you will find that your scripts have a lot of repeated code, or that you are spending a lot of time adjusting parameters at the top of the file
• Refactor out repeated code into function calls in your scripts and implement the function either in the same file or a separate one
• Be comfortable with the calling and return syntax of your programming language environment, whether it is bash or Python
• Don't repeat code in scripts, refactor them to functions

Group commonly used functions into libraries

• If you have to write a lot of software functions, consider designing and releasing a library so that others do not have to share your misfortune
• Check that nobody else has implemented the functionality you need
• If something close exists, it may be worth adapting to your needs if the project is of high quality and suitably licensed
• Openly licensed non-commercial libraries tend to have a much longer effective lifespan than unreleased codes
• Share your code with others, and use their code

Basic strategies

• Use languages and libraries that reduce the complexity of your work
• It is worth installing a complicated or expensive software tool if your computations or model are naturally expressed with it
• Always look for opportunities to write less code
  • you will have to do less initial work (sometimes)
  • you will introduce less bugs
  • your code will be easier to understand and maintain
• Keep individual functions short, single-purpose, possible to be confident in festooning.

Back up your data!

Use version control for checkpointing and collaboration

• use local version control software to checkpoint personal code development
• checkpointing your work encourages wild ideas and late-night coding sessions
• you can easily restore back in the morning if it was a bad idea
• use distributed version control to collaborate with others
• We advocate Git, but you may be stuck with whatever your group uses

Principles of verification and validation

• verification - is your code correctly written?
• Be paranoid.
• test small things!
  
• test that what you assume is TRUE is in fact so.
• test frameworks can help you verify your code
• validation - do your computations accurately model the phenomena in question?
• not a good candidate for automation. (Not sad at all)

Document your computational work

• Save every bit of code you use for generating publishable results
• Document and comment your code for yourself as if you will need to understand it in 6 months
• use README files liberally
• as well as file-level, function-level, and inline documentation
• If any piece of code is too complex to easily describe, consider refactoring it

You sometimes need geeks. You never need dorks.

Programming Languages for Scientific Computing

Matthew G. Knepley

Preprint: http://arxiv.org/pdf/1209.1711.pdf

Gives an overview of modern programming languages and techniques such as code generation, templates, and mixed-language designs. This is a preprint, so expect some rough spots.

Two Solitudes

Greg Wilson

Slides: http://www.slideshare.net/gvwilson/two-solitudes

Describes Greg's journey as a scientist and leader for the Software Carpentry project, provides some insight into the differences between industry and academics.

Best Practices for Scientific Computing

D. A. Aruliah, C. Titus Brown, Neil P. Chue Hong, Matt Davis, Richard T. Guy, Steven H. D. Haddock, Katy Huff, Ian Mitchell, Mark Plumbley, Ben Waugh, Ethan P. White, Greg Wilson, Paul Wilson

Preprint: http://arxiv.org/abs/1210.0530

Good summary paper of many fundamental practices for working with and developing scientific software. This is a preprint, so expect some rough spots.

What Every Computer Scientist Should Know About Floating-Point Arithmetic

David Golberg

Web article: http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html

Introduction to the IEEE floating-point standard, its implications, and many of the common pitfalls when using floating-point numbers in scientific computing

Science Code Manifesto

http://sciencecodemanifesto.org

Publicly signed commitment to clear licensing and curation of software associated with research publications.