Thread Status:
Not open for further replies.
  1. Steerpike
    Online

    Steerpike Felis amatus Supporter Contributor

    Joined:
    Jul 5, 2010
    Messages:
    11,059
    Likes Received:
    5,264
    Location:
    California, US

    Computing errors in science

    Discussion in 'Debate Room' started by Steerpike, Nov 19, 2015.

    Interesting article, here: http://f1000research.com/articles/3-303/

    The author points out the error rate in commercial-level software, not to mention in code that is written by PIs or graduate students to crunch data. One suggestion is that numbers should be run through multiple independent software programs to ensure that the results of the number crunching are reproducible and therefore less likely to be the result of software error.

    Of course, another issue with software for things like modeling is taking into account what presumptions are built into the model so you know how to evaluate what you're getting out of the program.
     
  2. thirdwind
    Offline

    thirdwind Contributing Member Contest Administrator Reviewer Contributor

    Joined:
    Jul 17, 2008
    Messages:
    7,351
    Likes Received:
    2,891
    Location:
    Boston
    I guess this is why it's still better if the person using the software has a relevant degree or some experience in the field. Anyone can use software with a little bit of training, but interpreting the results and making sure they make sense is a whole different animal. One of my engineer friends puts it best: a computer program is only as smart as the person using it.

    On a related note, a lot of programs have rounding errors and whatnot and still give meaningful answers. In fact, a lot of software used in engineering uses approximations, and we still rely on it to build bridges, planes, etc. This is something I have always found fascinating.
     
  3. 123456789
    Offline

    123456789 Contributing Member Contributor

    Joined:
    Jan 28, 2012
    Messages:
    6,336
    Likes Received:
    3,084
    If the simulated data doesn't fit the experimental data, try, try again....



    Because eventually it will.
     
Thread Status:
Not open for further replies.

Share This Page