tips

tipoftheweek

Tips of the (hopefully) week for research for my group at BU

Future ideas:

• Source code versioning
  • Why do it - show example
    • Useful for you; perhaps even more useful for your users
  • General idea and diffs or patches
  • Current practices - SVN and Git
  • Git on cluster
• dotfiles - bashrc and aliases
• Reproducibility
  • http://ropensci.org/blog/2014/06/09/reproducibility/
  • http://theconversation.com/science-is-in-a-reproducibility-crisis-how-do-we-resolve-it-16998
  • http://spectrum.ieee.org/robotics/artificial-intelligence/machinelearning-maestro-michael-jordan-on-the-delusions-of-big-data-and-other-huge-engineering-efforts
  • How much of the wood is woody?
    • WOW! Full analysis, from dataset to PDF
    • http://richfitz.github.io/wood/
  • MRV... why reproduce the map if we can just perform accuracy assessment and get bias-corrected area estimate?
    • In-fighting in the literature about definitions of trees, changes, carbon densities, etc.
    • More knowledge of behind the scenes --> less pithy arguments?
    • Small samples means small disagreements can result in large differences in estimates
• Remote utilities
  • nohup
  • screen
  • tmux
  • vnc
  • SSH-ing directly into SCC nodes:
    • http://superuser.com/questions/96489/ssh-tunnel-via-multiple-hops
    • http://superuser.com/questions/107679/forward-ssh-traffic-through-a-middle-machine
• Workflow management
  • Needs:
    • orchestrate tasks
    • graphs
    • logging
    • SGE - qsub stuff
  • Libraries:
    • Ruffus
      • Python decorators to tie functions/etc together
    • pyutilib.workflow
    • Spotify's luigi (not usable on SGE?)
    • COSMOS
    • doit
• 3rd party libraries
  • $H_0$: someone already wrote the thing you need
  • how to find it
  • how to search it
  • when is it not worth using
    • licensing
    • documentation
    • portability concerns
• Benchmarks and profiles - how to speed up your code
  • CPU vs IO
  • Memory profiling
  • Bottlenecks and our cluster
    • Speed ~ # jobs
• "Defensive programming"
  • User interfaces (command line)
  • Why check?
    • You can provide a clear message about what went wrong versus a stack trace
  • http://software-carpentry.org/v5/novice/python/05-defensive.html
• Testing
  • Test driven development
    • Good when you know the expected behavior
  • Regression testing
    • Especially useful when iterating on your algorithm
  • NOT for exploratory; but for code you will reuse
  • http://science.gsfc.nasa.gov/sed/content/uploadFiles/highlight_files/TDD_For_Scientists.pdf
• Databases
  • SQL language
  • In practice - sqlite
  • For geography - PostgreSQL with PostGIS
  • General knowledge - SQL vs noSQL / structured vs unstructured

Resources

• http://software-carpentry.org/lessons.html