6 ms·

In applied math, data science, statistics, the OP is right: free software is the norm.

by enriquto 7y ago

nfoz 7y ago
Really depends which niche you're in. Matlab for example is also a major player.
- RosanaAnaDana 7y ago
  Not in production environments. Push is (as it should be) towards open source.
  - kgwgk 7y ago
    Mathworks' revenue is $1bn, it has doubled in the last decade. Someone may be using Matlab.
    - Spivak 7y ago
      A ridiculously large number of people use Matlab as a tool. I can't think of a single engineer that doesn't curse its name regularly but I've yet to encounter any actual applications written in it. It's Excel for engineers.
      - gmueckl 7y ago
        I have seen people doing amazing and complex things with Matlab, including fancy GUIs and stuff. But it is almost exclusively used for internal tools as far as I know.
      - new4thaccount 7y ago
        I think this often happens anytime there is a large amount of legacy software written by folks that had software has a tertiary skill on their tool belt. The Matlab IDE is not bad though and is a lot like Python's Spyder IDE. Also, a lot of Matlab commands have fully browsable source provided I've been told.
LolWolf 7y ago
I’ll add to the list: mathematical modelling, optimization, control, essentially all machine learning, etc. Except for perhaps optimizers (which some of the best are closed source, e.g. Mosek), the entire rest of the stack is open source.
- kgwgk 7y ago
  CUDA is not open source.
  - LolWolf 7y ago
    Sure. Neither is the hardware, nor is Intel’s or the Motherboard’s, etc. Would it be ideal? Sure. Is it possible to have every level be completely open source? Maybe, but you’d struggle. I still believe the point is clear, though.
    - kgwgk 7y ago
      Ah, you're right! The whole stack is open source if you ignore the non-open-source parts. (Intel compilers are also popular, by the way.)
      - LolWolf 7y ago
        Look, the point here is to be as open as it is to be practical. In my lab, we attempt to use as much open source software and parts as is possible without literally spending years waiting for RISC-V to become mainstream along with fully open source GPUs and computing cores, open motherboards, RAMs, etc... this doesn't even include fabrication or other useful software which runs our machines (do you want to have a several-million-dollar photolithography machine running on poorly maintained OSS?). Yes, should that machine be open in a perfect world? Of course, that would be incredible and in the spirit of verifiability. But right now it is not and there is little we can do about it. It's clearly impossible (or completely and utterly impractical) to pretend that somehow everything can be done in a fully open source way, and it's easy to go down the rabbit hole pretending to be on the high horse of "our stuff is more open." My point is that what can be made with OSS, we almost always attempt to make with OSS, not everything we make is perfectly open and transparent, because it literally cannot be without so much effort as to render the entire thing completely useless. So, sure we use Mathematica in some things when the OSS alternative is mostly unworkable or unusable, but by and large we attempt to be as open as possible such that people can verify our results without needing to spend thousands of dollars on software whose internals we don't generally know. Do we rely on Intel software and hardware which could have another floating point bug? Sure. But it's hard to be at all productive if you don't at least assume a bare minimum, even if that standard is not completely open.
ras_west 7y ago
I think there are people in applied math data science and statistics who use Windows and OSX.
mruts 7y ago
Excel? SAS? Stata? Matlab? All are very popular tools in data science and statistics. You could argue that open source tools have become more popular recently, but I would hardly call open source software the "norm."

In applied math, data science, statistics, the OP is right: free software is the norm.

nfoz 7y ago

Really depends which niche you're in. Matlab for example is also a major player.

RosanaAnaDana 7y ago
Not in production environments. Push is (as it should be) towards open source.
- kgwgk 7y ago
  Mathworks' revenue is $1bn, it has doubled in the last decade. Someone may be using Matlab.
  - Spivak 7y ago
    A ridiculously large number of people use Matlab as a tool. I can't think of a single engineer that doesn't curse its name regularly but I've yet to encounter any actual applications written in it. It's Excel for engineers.
    - gmueckl 7y ago
      I have seen people doing amazing and complex things with Matlab, including fancy GUIs and stuff. But it is almost exclusively used for internal tools as far as I know.
    - new4thaccount 7y ago
      I think this often happens anytime there is a large amount of legacy software written by folks that had software has a tertiary skill on their tool belt. The Matlab IDE is not bad though and is a lot like Python's Spyder IDE. Also, a lot of Matlab commands have fully browsable source provided I've been told.

LolWolf 7y ago

I’ll add to the list: mathematical modelling, optimization, control, essentially all machine learning, etc. Except for perhaps optimizers (which some of the best are closed source, e.g. Mosek), the entire rest of the stack is open source.

kgwgk 7y ago
CUDA is not open source.
- LolWolf 7y ago
  Sure. Neither is the hardware, nor is Intel’s or the Motherboard’s, etc. Would it be ideal? Sure. Is it possible to have every level be completely open source? Maybe, but you’d struggle. I still believe the point is clear, though.
  - kgwgk 7y ago
    Ah, you're right! The whole stack is open source if you ignore the non-open-source parts. (Intel compilers are also popular, by the way.)
    - LolWolf 7y ago
      Look, the point here is to be as open as it is to be practical. In my lab, we attempt to use as much open source software and parts as is possible without literally spending years waiting for RISC-V to become mainstream along with fully open source GPUs and computing cores, open motherboards, RAMs, etc... this doesn't even include fabrication or other useful software which runs our machines (do you want to have a several-million-dollar photolithography machine running on poorly maintained OSS?). Yes, should that machine be open in a perfect world? Of course, that would be incredible and in the spirit of verifiability. But right now it is not and there is little we can do about it. It's clearly impossible (or completely and utterly impractical) to pretend that somehow everything can be done in a fully open source way, and it's easy to go down the rabbit hole pretending to be on the high horse of "our stuff is more open." My point is that what can be made with OSS, we almost always attempt to make with OSS, not everything we make is perfectly open and transparent, because it literally cannot be without so much effort as to render the entire thing completely useless. So, sure we use Mathematica in some things when the OSS alternative is mostly unworkable or unusable, but by and large we attempt to be as open as possible such that people can verify our results without needing to spend thousands of dollars on software whose internals we don't generally know. Do we rely on Intel software and hardware which could have another floating point bug? Sure. But it's hard to be at all productive if you don't at least assume a bare minimum, even if that standard is not completely open.

ras_west 7y ago

I think there are people in applied math data science and statistics who use Windows and OSX.

mruts 7y ago

Excel? SAS? Stata? Matlab? All are very popular tools in data science and statistics. You could argue that open source tools have become more popular recently, but I would hardly call open source software the "norm."