The 0 indexing in python does really and truly suck sometimes though.
Now that's something you usually don't hear. At my university I heard a bit of "urban legend" about how back in the day (think early 1990's) a couple of professors got a peer-reviewed article published because they found out that Matlabs 1-indexed implementation of some algorithm resulted in numerical errors, which they measured and corrected. Don't really remember the details, and most of those involved have already retired.
The problem isn't whether 0 or 1 are "right" or not, it's the inconsistency. It makes transcribing something from a textbook harder, because the indexing logic in a textbook algorithm can get quite intricate. It's even worse if they use slices like M[i:j,m:n].
Indexing from 1 is the standard in many areas, going back many decades. SWEs have adopted a different convention.
It is inevitable to have to use both conventions if you do varied math stuff. Thus, you will never be happy.
So you end up having to translate from one-indexed derivations to zero-indexed code and it leads to different bugs. Ultimately zero-indexing just trades one layer of bugs for another.
IMHO zero-indexing makes sense to people thinking about the machine, not necessarily the work people are trying to use the machine to do.
To be fair sometimes the opposite is true and zero-indexing gives less complex expressions.
How about I drop a link about Einstein Notation with a curt observation that there are good reasons it uses both zero and one indexing simultaneously?
As I said it depends on how you are thinking about things. Indexing makes sense to people who are focusing on the machine.
"The purpose of computing is insight, not numbers." (Hamming)
Machines and notation are both tools. Aesthetics do matter particularly for tools used for thought.
There is no correct answer and imposing one tool's limitations on everything is not useful. Particularly not "just because Dijkstra said so".
Sometimes 0-indexing works well. Sometimes it sucks. Sometimes 1-indexing works well, sometimes it sucks.
I did in fact give you a reference. The fact that it's irrelevant to you and dismissed as merely "aesthetics" and that you think a computer scientist is the domain expert of all science tells me everything that needs to be said--you never deal with questions of correctness beyond mere transcription.
What I said is that moving from Octave/Matlab to python's zero indexing sometimes does truly suck. And I mean it. That doesn't mean 1-indexing never sucks or that I am inaware of Dijkstra's opinions. Having to arange(1, N+1) or 1:N+1 or whatever everywhere isn't eliminating off-by-one errors.
Matlab and Octave and FORTRAN etc are domain languages designed for linear algebra. They are of course going to be more convenient for their domains.
Furthermore it depends on what you're doing with that first centimeter. In many instances the first centimeter is correctly considered to be 0.5
I find it easier not to make mistakes when I index from 1.
The issue stems from the word "index" in the sense of "to count" .. that the first cell is numbered 1 makes sense to some.
What is actually happening is measurement, or the offset from the origin of memory.
The first inch starts at zero on the tape measure and ends at 1, that first inch has an offset of zero.
For all manner of reasons to do with measuring memory as one might measure wood a great many people think in terms of memory offsets.
So every time I see `for i in range(10)`, I have to do the mental computation: I am repeating this operation on numbers from 0,...,9.
I would rather write `for i in range(9)` and know that I am operating on 0,..,9.
Exclusive upper bounds mean slices work neatly: a[0:10], a[10:20], a[20:30] etc. That's very neatly represented in a loop from zero in steps of ten. That pattern turns up all over the place when accessing arrays.
I've spent enough time in MATLAB to have been scarred by that kind of array slicing and off-by-one errors.
In many non-CS engineering disciplines Octave (and Matlab) is quite common. For doing everyday numerical calculations. Python and Julia do not compete with the simplicity of using GNU Octave.
Python I understand. The numpy notation sucks big time for translating straighforward notation of engineering disciplines and is a huge step backwards in readability; but Julia? That's a surprise. In fact the opening code in the Octave page can be copied almost exactly into Julia and it works.
It is really costly for a dept. to switch to Julia, say, as you would need all your colleagues (I teach applied maths and use matlab/octave) to make the switch lest you annoy the students (because ones learn one thing, others other).
I recently found Spyder IDE for Python which feels a lot like MATLAB. Despite my hate for MATLAB, I did at times appreciate the layout of the IDE and the way variables are still available after running a script. Spyder fortunately has a similar interface, except even better since you can restart the interpreter without restarting the IDE. I highly recommend it as a MATLAB replacement!
But truth be said, nothing so far quite managed to replicate Matlab syntax succinctness and ease - even if it’s a (unstated?) design goal of Julia.
- Fast, in an area where speed is important
- Can be made faster for repetitive tasks by communicating with a GPU using Vulkan, CUDA etc
- Can incorporate a GUI, and 3D graphics, which helps for exploring things visually, tweaking parameters, doing time simulations etc.
- Can share. Ie, I showed my brother yesterday by sharing a 10Mb executable. Matlab has license complications and you need it installed. Sharing Python programs is a disaster. (Docker? Pip? Pyenv? Virtualenv? System Python? System dependencies for the C dependencies?)Python is the big one, all of the aforementioned chemists are either intermediate or advanced in that. The runner-up seems to be Julia, which I personally have no experience with. The big guys are Fortran and C++. I prefer C for tasks of this nature, but I also shill Scheme so don't listen to my opinions on programming languages.
Best of luck on your computational chemistry endeavours!
The same would not be true if we were supposed to teach them C++ for example. It's a huge, sprawling language, full of difficult concepts that are unfriendly to teach, and equally filled with foot guns that if you don't teach them will definitely maim your students, but if you do teach them take up yet more precious time on minutiae.
Safe Rust wouldn't be as hard to teach as C++ but it's no picnic. So even if the Chemists decided that ideally they'd like their undergraduates to learn Rust instead of Python, I think the argument would be that it can't be done on the existing timeline.
Compared to Perl, sure, Python is teachable. But I'm willing to speculate that Safe Rust as a semi-pure functional language (basically: when in doubt, .clone() all the things and don't even think about the minor hit to performance) can also be taught as a first-time programming language. Rust may be at a disadvantage to Python wrt. bindings to the existing ecosystem of C/C++/F0RTRAN scientific codes, but that's a temporary state of things.
> but then they're going to hit the large legacy of existing code and wonder why you taught them something that not related to what they're going to use.
OTOH other languages should be easier to learn by reference to a comparatively elegant language like Rust. Python, Fortran, C++ etc. have almost nothing in common but Rust could arguably be a good introductory baseline that shares features with all of these.
That is a very good question.
IIRC at the time Perl had a bad reputation because of the weird syntax "$%@" (not weird to shell programmers, so much, but they were weird...)
Python was "object orientated" and from a computer scientist.
Perl is very imperative and informal. Python felt to people like a real language, Perl did not.
I am reaching for reasons as I was very disappointed that Python was the winner. I have never liked Python (personal taste - meaningful white space? Really?) and knew Perl very well.
(Perl Vs. PHP is a different, much more interesting, and tragic story)
Many people conflate their own irritation from hearing about the thing X often and against their will, with the quality of the thing X.
IMHO critical thinking and maturity demands that we make that distinction.
I have worked several times in Rust companies and their devs are extremely hardcore and very quiet. You couldn't tell these people are responsible for the technical operations and code of worldwide financial services.
So again, be a bit more charitable. Every community has its rotten apples. That says nothing about the quality of the thing that the community stands for.
Don't get me wrong, insufferable people still exist! But I no longer judge an entire community by them.
Or for simulating molecular motion, draw it out in 3D, with 6DOF FPS-style camera controls etc.
I imagine Pandas would be OOM too slow for this. Compared to Python, Pandas is, in my experience far too slow. Another speed limitation is Matplotlib's 3D plots make my computer crawl when rotating them etc; and they're not very interactive, at least on their own.
When dealing with these systems, a particular computational challenge is that they're 3D, so computation scales with precision^3. (This also adds complications to visualization, as I alluded to above)
* Fast - in many cases faster than Rust, although the difference is inconsequential relative to Python-to-Rust improvement I guess.
* _Really_ utilize CUDA, OpenCL, Vulcan etc. Specifically, Rust GPU is limited in its supported features, see: https://github.com/Rust-GPU/Rust-CUDA/blob/master/guide/src/... ...
* Host-side use of CUDA is at least as nice, and probably nicer, than what you'll get with Rust. That is, provided you use my own Modern C++ wrappers for the CUDA APIs: https://github.com/eyalroz/cuda-api-wrappers/ :-) ... sorry for the shameless self-plug.
* ... which brings me to another point: Richer offering of libraries for various needs than Rust, for you to possibly utilize.
* Easier to share than Rust. A target system is less likely to have an appropriate version of Rust and the surrounding ecosystem.
There are downsides, of course, but I was just applying your criteria.
Would also def benefit from the larger selection of C++ libs, especially for GUI and graphics.
Given the main computation issue is doing the same operation on many values in a 3D array, using the GPU more would be great. And my current setup for that is clumsy.
I disagree on easier to share. Binaries work the same either way. Compiling from source on rust is generally install Rustc, and do `cargo b --release`. Compiling someone else's code in C++ is... complicated.
Yes, that's possible with a combination of:
* Unified address space for pointers (and C++ references are basically pointers)
* Paged virtual memory on GPUs
... however, remember that many/most data structures which make sense on the CPU may not make sense to use on a GPU. And - NVIDIA very often implements and promotes certain features because they sound good in marketing, not because they're actually that useful. Or because they let you make a super-slow program into a meh-speed program, not because you would use them in a really-fast program.
> Compiling from source on rust is generally install Rustc, and do `cargo b --release`.
Well, I guess you may have a point, but let me still make a couple more arguments.
First, installing Rust (and related libraries/tools/whatever) is not trivial for people who don't know rust, and some OS distributions only offer you an older version of rust using their distro-package-managers. It's more likely that a C++ environment is already set up on someone's machine... TBH, though, if the distro is older, that might not be a perfect fit for what you're building.
Second C++ package managers becoming more popular, e.g. Conan: https://conan.io/ ; and that helps for when your OS distro doesn't cover you.
> Compiling someone else's code in C++ is... complicated.
This has actually improved a whole lot over the past... say, decade or so!
* Unzip/untar
* Configure the build with CMake (cross-platform build system generator - almost ubiquitous these days in C++ projects): `cmake -B build_dir/`
* Assuming you have the dependencies set up - it should "just build": `cmake --build build_dir`
* Run the thing from the build directory or install it with `cmake --install build_dir`.
... and if you want to tweak package options more easily than via the command-line, you can use a TUI (ccmake) or GUI (cmake-gui).
The problems start when you're missing dependencies, or if the author didn't make their code platform-independent / multi-platform.
Do you have any examples? AFAIK properly tuned rust and c++ will perform largely the same. Actually, Rust should have a bit of an edge due to the prohibition of aliasing. In practice it can vary if a standard library implementation is suboptimal or the compiler has suboptimal codegen into Bitcode but that’s generally going to be rare these days I think.
> Richer offering of libraries for various needs than Rust, for you to possibly utilize.
Are you talking for computational chemistry specifically or general libraries?. Don’t know about the former. For the latter I’ve found not only are more interesting libraries available, there seem to be generally high quality versions. Additionally, “cargo add xxx” is infinitely faster than integrating some random third party c++ dependency. Not to mention that C++ falls on the floor if one dependency requires transitive dependency X and another (or you) requires X at an incompatible version? Rust handles that elegantly where two different modules can depend on the same package at different versions without conflict without you needing to worry about it beyond the size of the executable.
> Easier to share than Rust. A target system is less likely to have an appropriate version of Rust and the surrounding ecosystem.
Again, haven’t that found to be my experience. Rustup lets you obtain any version of the rust tool chain, no muss no fuss. Additionally, it does a phenomenal job of not worrying about which compiler vendor you’re using (there’s only one) and more importantly there’s no versioning issues with libraries written against an older version of the language - 2021 and 2018. Oh, if you’re on windows, you also don’t have to worry about whether the target person has the right c++ runtime library version installed. With rust they just run the exe.
I wonder what's an equivalent of Eigen? https://eigen.tuxfamily.org/index.php?title=Main_Page
> if you’re on windows, you also don’t have to worry about whether the target person has the right c++ runtime library version installed
It's usually easy to build such binaries for Windows. That's a single combobox in project settings. Here's an example, just running the .exe works: https://github.com/Const-me/Whisper
2 seconds on Google yields:
https://rust-lang-nursery.github.io/rust-cookbook/science/ma...
https://docs.rs/GSL/latest/rgsl/
I have no idea the quality but you’ve also got bindings to more mature libraries:
https://docs.rs/lapack/0.14.1/lapack/
And https://www.erikpartridge.com/2019-03/rust-ml-simd-blas-lapa... is an overview from t years ago and does say it’s maybe not as good although not sure how much has changed in that time (eg has nalgebra matured sufficiently?)
It’s possible that Rust is lacking in good linear algebra libraries. I’m not as familiar with that space.
In principle, Rust has some overhead for being "safe", which C++ does not. Also, C++ benefits for longer period of time and more people working on optimizing its libraries and compilers.
> Are you talking for computational chemistry specifically or general libraries?
Ah, indeed, no. I have no idea which chemistry libraries are available for Rust and for C++. I was referring to scientific computing, algebra and other heavy-lifting work.
> Additionally, “cargo add xxx” is infinitely faster than integrating some random third party c++ dependency.
That's not one of OP's criteria. But - seeing how you like package managers, try `conan install /path/to/srcdir` from the build directory. For more details, read:
https://docs.conan.io/en/latest/getting_started.html
> Not to mention that C++ falls on the floor if one dependency requires transitive dependency X and another (or you) requires X at an incompatible version?
C++ is a language, it doesn't have dependency management. As for libraries, that situation is so rare - in my experience - that a chemistry person should not even be bothered to consider it.
> haven’t that found to be my experience.
Well, I learned something new today... I didn't know about rustup. Was it introduced recently?
Do you know of any specific safety features that makes rust slower? About the only one I can think of at runtime is that everything is bounds checked but that’s usually fixable in many ways (and technically c++ has that too except it’s applied randomly based on subtitle API differences like [] isn’t but .at is). As for quality of the standard library it’s hard to say. I haven’t found it to be bad. The HashMap implementation is better out of the box, a Vec/vector doesn’t really have anything special. File system APIs, error handling, modules, and async all feel more mature and polished (not performance per se but just overall completeness of the language). What is on your mind when you say that the c++ standard library is higher quality? Oh, and which of the three major implementations are you referring to?
The compilers story is actually even better because very little optimization happens in the frontend and thus a lot of the optimizations done for c++ apply to rust too. One specific way the language itself is better is that it disallowed aliasing so it’s closer to Fortran for numerical stuff and there’s various optimizations that c++ just can’t do. If I recall correctly that hasn’t been fully hooked up to the compiler due to bugs in LLVM but that will get resolved eventually. Overall, I’m not familiar with the claim that Rust code compiles worse than c++. About the only thing I’ve run into so far is bound checking showing up in hotspot code that I wrote non idiomatically because I’m still new to the language.
> Well, I learned something new today... I didn't know about rustup. Was it introduced recently?
1.0 is 6 years ago so not recent at all (not sure if there were pre 1.0 releases too): https://www.reddit.com/r/rust/comments/5iqsmg/rustup_100_is_...
There's nothing like octave (or MATLAB) for scripting to "get shit done." There is a plethora of packages and tools to do whatever you need.
Need to import excel data and graph it? Octave is great. Need to connect to an sql database, pull data and export it? Octave. Linear algebra. Nuff said.
It's a great tool for engineers / scientists who don't want to "program" (although you can!) and want readable syntax to just git r dun. Also, there's a shit ton of m code floating out there on the internet and in engineering schools / university research groups. Octave isn't 100% m compatible but if you don't have a special tool boxes & functions you can get them to run in octave with a little debugging.
I love octave. It's gotten me out of a lot of binds.
I also wish the state of science/engineering software shook out differently. There's plenty of money to pay Mathworks. Is there some kind of license like pay us if you're doing commercial work or publishing research on grants worth over $XXX, otherwise consider it open source?