• Edd (unregistered)

    time.sleep is so that the HPC can see the debug output before it scrolls. They didn't remove all of it, as seen by the +++ lines.

  • (nodebb)

    You see, the code is bad, the comments are meaningless, the wage is unjustifiably high, but it's MACHINE LEARNING CLOUD DATA OCEAN!!! All the failures are forgiven.

  • Steve (unregistered) in reply to Mr. TA

    MACHINE LEARNING CLOUD DATA OCEAN!!!

    DevOps AI blockchain NFT

    That's worth a line on anyone's buzzword bingo card, surely

  • Prime Mover (unregistered)

    time.sleep is clearly so the computer can take a breather and sit down for a few seconds so it doesn't get tired. After all, that's what my doctor told me to do when I'm walking round the garden for my daily exercise. That's why I've got plenty benches.

  • 516052 (unregistered) in reply to Mr. TA

    The only wage that is too high is the one we are jelous off.

  • Prime Mover (unregistered)

    Oh yes, and of course TRWTF is that Noreen can't spell "ad nauseam".

  • RLB (unregistered)

    The problem, I suspect, is that ML people aren't really programmers. They're (some kind of) scientists. And, well, the average code quality of the average scientist...

  • my name is missing (unregistered)

    Maybe the machine learned something, but these highly paid contractors clearly did not.

  • (nodebb)

    What? The number given as an argument for time.sleep(5) is interpreted as seconds instead of milliseconds? Though I prefer to use a Timespan(C#), my C favorising cow-orkers insist on their primitive obesity ehm obsession, and ... yeah, I hope we won't get to ML and Python in our shop then!

  • Sigako (unregistered) in reply to RLB

    time.sleep's are a job security - if they have nothing to report for the next patch, they remove one of them and claim drastic increase in speed. It's a common practice.

  • Sigako (unregistered) in reply to RLB

    Legitimate scientists, no matter how bad the code they write, don't want to have anything in common with these losers for the most part either. Basically, they cannot do science, cannot write code and thus pretend they can do both.

  • Sole Purpose Of Visit (unregistered) in reply to RLB

    Well, no. That is not the problem.

    Let us stipulate that the ML guy wrote the model correctly. Now, it doesn't matter whether this model is based on R, or on any one of Google's various map-reduce cloud offerings, or on Microsoft's Scope (the one I know best). The end result is going to be a data-set. Job done, $300 per hour, thankyou very much. Beyond that, the ML specialist doesn't really have to know anything very much about programming, much less a python script to convert to a PDF'

    That is why you pay a "specialist" to write a python script to convert the data-set to PDF.

    There may well be a contractual snafu here, whereby for no good reason the company entrusts the ML guy to sub-contract the python script out to some random dweeb, but the worst you can say about the ML guy is that he picked a moron to do the job.

    The actual WTF is that, at all points from contract to requirements to testing to delivery, nobody checked the quality of the python script. I say "specialist" up above, but let's be honest. Any of us could do that job better.

    Presumably the data-set is still available to any future programmer. All the company needs to do is to hire a very clever amoeba at, say, $100 per hour to write the necessary python script.

    I am happy to volunteer to be that very clever amoeba.

  • Foo AKA Fooo (unregistered)

    /* printf () Hello + World""? // TODO: FIX write {} => 'stdout' = FILE_NOT_FOUND */

    Any bets on the payload size, i.e. number of lines that do some actual work? Based on what we've seen to far, my bet will be a single figure.

  • Flips (unregistered) in reply to Sole Purpose Of Visit

    I can confirm you are a very clever amoeba.

  • LCrawford (unregistered)

    Artificial Intelligence at its finest!

  • Sole Purpose Of Visit (unregistered) in reply to Flips

    Obviously I should reproduce then. Watch this!

    Christ, that hurt. Plus which my IQ seems to have dropped by 50%.

    Never mind. I can now offer my services as a PHP consultant (last worked with PHP5), using templates to output the PDF. Having doubled myself, my rates are now a very reasonable $50 per hour.

  • (nodebb)

    Physics PhD with extensive CS background on top here. Judging from my current programming job in a science-type software, I can confirm the statements mostly.

    It is probably easier to work on such code, if you don't know enough to realize how much is being done wrong...

    Addendum 2021-05-12 11:24: It also doesn't do anyone any favors when programming classes for non-CS scientist/engineers are taught by people of their own field. If you teach anyone programming, please also make them do a basic Software Engineering lecture (topics like technical debt, coding practices, why-you-dont-want-to-use-fortran), data structures/algorithms, and maybe something on database design (a lot of that translates directly to coding practices, e.g. the consistency issues arising from duplicating data).

    It is a hard call to fit it into a curriculum, but otherwise it's better to have people stick to Matlab/Mathematica. No sense to teach someone C++ of if you're not going to teach them about data structures and algorithms.

  • MiserableOldGit (unregistered) in reply to R3D3

    Well as a grumpy engineer from a profession that earned that title (before yours!) I will both agree and disagree with your rant. We all know engineers get their programming lectures from the guy in the Maths department who was drunk enough not to say "no, not me" at the right moment.

    I've no idea what the answer is, but, unfortunately, it's not here.

  • X-Centric IT Solutions (unregistered)
    Comment held for moderation.
  • 516052 (unregistered)

    I just wish maths people would stop thinking they know programming. They don't. Mathematics has about as much to do with programming as theoretical phisics has with architectural engineering.

  • (nodebb)

    Some mathematicians know programming. They're very much in a minority. Indeed, the skills required are very different: much of math is about finding clever shortcuts so that you can avoid diving deep into the complexity in the first place, whereas much of programming is about finding ways to provide all the detail to the very stupid machine while not going completely crazy in the process. They're similar if don't look carefully, but that similarity is skin deep; math thrives on assumptions, whereas programming really has to spell things out in the end, and assumptions need to be handled pretty rigorously or things really go wrong.

  • tbo (unregistered) in reply to 516052

    Seems to me that programming at least has a lot in common with boolean math.

  • Neveranulll (unregistered)

    It seems that the only important skill shown here is the ability to negotiate for $300 or $200 an hour. I guess I wasted my time concentrating on continually improving my software development skills, only to continually get emails and calls offering “lucrative long term contracts” for $57 to $70 an hour.

  • Mooo (unregistered)

    ML is overrated? No quack!

  • (nodebb) in reply to dkf

    math thrives on assumptions

    Wait what?

  • King of the Sun (unregistered)

    There's a program called DVC, Data Version Control. DVC will upload your ML data sets to online file storage, such as Google Drive. If you upload a CSV file to Google Drive, Google Sheets opens by default. Google Sheets has a "Save to PDF" feature, as well as the ability to copy a live range of data to Google Docs. Google Docs also has a "Save to PDF" feature.

    yea. That sounds like something a scientist under contract can handle.

  • King of the Sun (unregistered)

    There's a program called DVC, Data Version Control. DVC will upload your ML data sets to online file storage, such as Google Drive. If you upload a CSV file to Google Drive, Google Sheets opens by default. Google Sheets has a "Save to PDF" feature, as well as the ability to copy a live range of data to Google Docs. Google Docs also has a "Save to PDF" feature.

    yea. That sounds like something a scientist under contract can handle.

  • Alexander Carlos (google)
    Comment held for moderation.
  • Alexander Carlos (google)
    Comment held for moderation.
  • David Mårtensson (unregistered) in reply to Sigako

    A speedup loop ;)

  • Some Ed (unregistered) in reply to jkshapiro

    To be fair, I do not recall ever seeing a proof for

    1 == 1

    I personally consider 1 + 1 = 2 to be contained within the definition of the number system, and thus not require a proof. But, as I recall, for millennia, this was considered a mathematical assumption, until Bertrand Russell came along and produced a proof for it. But, if I recall correctly, even Bertrand Russell assumed that 1 was and is and will always be equal to 1... along with 6 other things I found so absurd that I could not remember them, because I'd always second-guess my memory too much for the thoughts to stick around long enough for me to commit them long term.

Leave a comment on “A MLearning Process”

Log In or post as a guest

Replying to comment #:

« Return to Article