/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality.

LynxChan updated to 2.5.7, let me know whether there are any issues (admin at j dot w).


Reports of my death have been greatly overestimiste.

Still trying to get done with some IRL work, but should be able to update some stuff soon.

#WEALWAYSWIN

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB

More

(used to delete files and postings)


Welcome to /robowaifu/, the exotic AI tavern where intrepid adventurers gather to swap loot & old war stories...


Selecting a Programming Language Robowaifu Technician 09/11/2019 (Wed) 13:07:45 No.128
What programming language would suit us and our waifus best? For those of us with limited experience programming, it's a daunting question.
Would a language with a rigid structure be best?
Do we want an object-oriented language?
How much do you care about wether or not a given language is commonly used and widespread?
What the fuck does all that terminology mean?
Is LISP just a meme, or will it save us all?

In this thread, we will discuss these questions and more so those of us who aren't already settled into a language can find our way.
>>130
This is good advice. I'd modify it slightly to say that a true novice should cut their teeth on Python (I hear that Python The Hard Way is a good primer) and then add some C on embedded systems - preferably something real-time - before touching C++. Python, "magic" and all, has the main advantage of laying down fundamental patterns of thought, and time-sensitive embedded C gets you close to the machine so that Python's abstractions won't damage long-term development. All learning consists of telling lies to small children; the trick is knowing when to take the lies away and replace them with the next, slightly less untrue set. In any case don't learn with JavaScript and definitely don't touch PHP until you have several better languages under your belt. I've observed that people who start with either of those two languages seem to suffer some kind of brain damage that carries over insufferably into other languages.

Clojure is also a decent choice as a first functional language. Functional experience pays dividends in any language. I hear that "Clojure for the Brave and True" is a good primer.

Don't neglect your tradecraft. Learn version control early. Understand how to collaborate with your future and past selves. Learn what good, disciplined thought feels and smells like and what sloppy thought feels and smells like. The Jargon File is a decent source of tips on developing your mindset, although some of it must be understood in its historical context.
>>428
>All learning consists of telling lies to small children; the trick is knowing when to take the lies away and replace them with the next, slightly less untrue set.
heh, creative way to word that anon. Great advice overall. What can I say, I love C++. I make no bones about it tbh. I learned C first, and wish I hadn't. If I'd had Ch27 of Stroustrup's PPP2 back then I would have both understood C better, and would have much more quickly transitioned to C++ thereafter.

Anyway, thanks for the advice and the tips. That kind of post helps out here a lot.
>>428
>The Jargon File
Thanks anon.
>>134
Could you please tell me the browser and OS you made this post on? Thanks.
>>539
Chromium derivative & Ubuntu.
>>428
It's funny, but I keep rethinking back to your words anon, especially the "Understand how to collaborate with your future and past selves." bit. That idea is helping to guide me in how to choose better function and variable naming, how to refactor code out to stay focused in the 'inner core' of a program, and generally how to do better architecture of my stuff. Thanks for the tip anon!
>>1271
I am glad that the words helped you, but don’t thank me, Anon - thank Master Foo: http://www.catb.org/~esr/writings/unix-koans/prodigy.html

>“Moving in accordance with the law of nature, it unfolds inexorably in the minds of programmers, assimilating designs to its own nature. All software that would compete with it must become like to it; empty, empty, profoundly empty, perfectly void, hail!”
Open file (21.65 KB 265x400 106728.jpg)
>>1272
very nice, here's one for you anon (though regrettably you'd probably have to get the 'made-from-atoms' version to enjoy it.

www.goodreads.com/book/show/106728.The_Timeless_Way_of_Building
Some lisp and AI related books:
>Structure and Interpretation of Computer Programs (SICP)
Book: https://mitpress.mit.edu/sites/default/files/sicp/full-text/book/book.html
Lectures: https://www.youtube.com/watch?v=-J_xL4IGhJA&list=PLE18841CABEA24090&index=1

>Paradigms of Artificial Intelligence Programming
Book and code: https://github.com/norvig/paip-lisp

>Practical Common Lisp
Book and code: http://www.gigamonkeys.com/book/

>The Common Lisp Cookbook
Book: https://lispcookbook.github.io/cl-cookbook/

>Clojure For The Brave and True
Book: https://www.braveclojure.com/clojure-for-the-brave-and-true/

>>128
>Is LISP just a meme, or will it save us all?
Lisp is good for learning (Scheme and SICP) and it allows one to build working programs fast (Clojure and Common Lisp), in addition, lisp has been traditionally used in AI programming.

>How much do you care about wether or not a given language is commonly used and widespread?
It's important that the language...
1) is tested and that it actually works
2) has a community, so you can get libraries, tools, learning resources and help
lisp (Common Lisp and Clojure), java and C++ are examples of languages that fulfill these requirements.

>What the fuck does all that terminology mean?
Object-oriented programming (OOP) just means that the language support one particular way of programming. nowadays, OOP is most widely used in the industry but functional programming is gaining a lot attention lately.
>>1322
Thanks for the links, Lisper-Anon.
This is an open-ended question so feel free to respond with your own opinion or thoughts.

As I understand it, compilers are continuing to improve to the point where even very savvy coders have a hard time beating them. So if a higher level programming language still compiles to fast code, then what are some advantages of using a lower level language in the context of AI and robotics?
>>1515
Well, I'm no expert in compiler design, and for the men who are, the phrase 'higher level programming language' applies to C and basically everything else outside of assembler or actual machine code.

You're making what is a mistake atp of just lumping everything into black and white. 'Very savvy coders' can probably implement a 'better' generic programming implementation than say, C++ template meta-programming by hand-writing assembler, but that would come with a few trade-offs;
-1. It would only be for a specific example, whereas the high-level meta-programming approach is fully generic.
-2. It would require months of the developers concentrated effort to produce a more performant version than say, GCC right out of the box.
-3. It would only work on a specific type of CPU, for example Intel or ARM.
-4. It would be a bitch to maintain that code after the fact (IMO).

So yea higher-level languages like C++ are certainly 'better' than low-level languages in that case.

Further, I'll presume the question is related to the common perception of 'higher-level' and please define that explicitly w/o using 'well, kinda like Haskell, you know' heh. While there are literally millions of professionals programming in C++ daily for a living, most amateurs consider it some a small niche thing (and therefore automatically both difficult and not worth learning).

Python certainly has more popularity and is considered higher-level. Examples of it's use abound in the sciences. But you certainly wouldn't write a 120fps FPS using it. In engineering, everything is a trade-off, and there's no free lunch--only elegant and inelegant approaches.

I use Python and C++ specifically because of OP's topic ITT: What programming language would suit us and our [robo]waifus best?

TensorFlow is arguably the single most import AI ML/DL framework right now. There's rather a good user API for the tool using Python. But the engine itself isn't written in Python, but rather in C++ (there's also an API in that language too ofc). TF performs metric shittons of intensive math operations and you definitely wouldn't be too happy with the performance if those were done in Python instead of C++. Yet calling into the C++ code w/ Python is a great idea.

Each language has it's place and there are definitely still quite wide performance gaps between very mature, highly-optimizable languages like C, C++, and FORTRAN, and modern coffee languages like Python, JavaScript, and C#, when performing math-intensive operations like performing tensor calculus on billion-cell matrices.
I don't think it's a good idea to advice beginners today to learn C++ as a start, bc its seems to be hard to work with. Python is what many scientists and beginners use. We'll need to glue different existing programs together, we won't write them completely on our own. For that, Python is the way to go. Speed only matters in some cases, btw.
>>4338 I understand your point Anon, but honestly, C++ isn't that hard to learn if you focus on the proper essentials. The problem is there's such yuge boatload of terrible, terrible educational material on C++ out there. It's really given guys the wrong impression of the language imo. I hope to correct some of that with the TDD. >Speed only matters in some cases, btw. Maybe in a general-purpose app or program. For practically the entire spectrum of software engineering related to robowaifus, speed is king. We literally will not be able to solve this domain successfully without well-written C & C++ programming. But yes, once the core engines are written and debugged, then adding in Python scripting on top of that will be suitable ofc.
>>4358 >I hope to correct some of that with the RDD.* derp. >>3001
>>4358 The engines for everything I can think of are general purpose, other people are using those and are working on improvements. Could you give me any example where we need fast software which isn't used outside of this project? Btw, there are ways to make Python code faster if necessary, same for many Lisp variants. However, I have no intenion to start one of these famous fights about programming languages here. I never tried CPP really, had no reason so far.
Open file (683.26 KB 1511x1997 fig-runtime-arch.jpg)
>>4376 >Could you give me any example where we need fast software Two words, Anon: Hard Real-time OK maybe that's 3 words heh There is literally nothing that is software controlled inside a safety-critical system (such as a Gynoid Robot expected to be around naive humans like children, for example) that isn't absolutely dependent on (billions of) calculations being done not only properly, but within a hard-limit time budget. I can go on about this for a while, but to save some time, do a little research of your own on those two words first Anon, and then come back and ask further ITT.
>>128 Good day fellow robo-lover! In reality, if we consider waifus to be both hardware and software, then there will be several groups of languages used: -Low-level languages such as architecture-specific assembly, C, maybe (unlikely) C++, which will be needed for embedded controllers with limited RAM and processing power. -Languages that will be running within an OS environment for machine-learning, bigger data proccessing etc. Those could be C++, Python, Go...you name it XD Coming from an electronics background, I think the topology of the waifu bot should be: -Simple, minimum driver code for interfacing to the hardware which by itself can execute a set of actions/steps -Higher level system for managing the overall state of the bot by aggregating data via APIs -I'm not fond of IoT, so I'd prefer the network connectivity to either not be present, or disconnected from the hardware control. I'll always recommend starting with C, just because it forces you to study the underlying architecture. However you should learn from a guide/book (like Ken and Ritchie's C) as learning it on your own is more frustrating without enough background. If you want to focus on higher level, I'm not sure what to offer. I use Python, but as mentioned earlier, it hides a lot from the beginner.
Open file (15.96 KB 260x343 knr_c_2e.jpg)
>>4403 >However you should learn from a guide/book (like [Kernighan] and Ritchie's C) Here's the bookcover. You literally cannot be involved in the C community w/o running into this.
>>4405 Hehe It's a well written book (though I've only read the first two chapters and usually use the others for reference). Also, perhaps you chaps might point me to the right thread, as this a little off-topic. If there's no thread I could create one. I wonder if anyone considered using alternative CPU architectures (i.e. not x86/amd64 or Arm) to make sure the waifu bot runs on an open architecture where the documentation can be readily acquired. For me this is important from a security perspective (Intel ME...) because a waifu bot is a very personal object and may be able to acquire a lot of info about you. I was once hyped about RISCV until I heard some open source projects struggling to get the necessary changes added to the standard (Libre-SoC). OpenPOWER might be interesting as it is well established.
>>4406 We are very concerned about the botnet issues here, both from security and privacy perspectives. We don't have a specified thread, but it is touched on in our SBC & microcontrollers thread. >>16 Please feel free to start an entirely new thread on the topic Anon if you feel qualified. Since this is such a technical topic, if you wouldn't mind make it an effort-post OP, detailing both the hazards of commodity processors, maybe something about the efforts in the alternative communities so far. Plenty of links always help round out a good OP ofc. Thank you.
>>4407 Sure, I'm quite busy on weekdays, but I'll see if I can do a little write-up for the weekend. I'm glad I found this place, as for once I can have a serious conversation about implementing waifuswith current technology, not just wishing and hoping! Thanks!
>>4406 BTW, this might be a good post to introduce to newcomers. >>2701
>>4408 >Thanks! You're welcome. Glad to have you here.
>related xpost >>6712
Here's something that might be useful. It seems to be picking up a lot of adherents. Nim https://nim-lang.org/ It's supposed to be easy to use like python but has a lot of advanced features without so much complication. Some features. It has macros which can make programs rewrite themselves according to situations and is supposed to raise productivity if you know how to use them. Macros are a big deal in LISP and C++. It can compile itself or to C or javascript. It has more than one level of garbage protection and memory protection so you can dash off stuff to test then advance speed by allocating memory yourself if what you wrote works well. Seems like something easier to use than C++ and LISP but more useful than python without being so big.
>>8524 Python is very common in science and education, though. It also has a lot of libraries, and speed isn't always a matter. However, I wanted to try out NIM for a while now. If it becomes popular here, I'm quite sure I will do so.
One thing is that if we're going to have different languages interacting then the language that will make this happen is C. Additionally, robowaifus are cutting edge tech, so we won't be able to cut ourselves some slack with convenient off the shelf hardware and operating systems, we'll have to deal with purpose-built hardware and software which inevitably comes with odd behaviors and optimizations, and to create and make use of those knowing the theory behind assembly is required, so that when the hardware pops up an assembly programmer will already know all about memory models and vector instructions and whatnot and be able to pick up programming for this new hardware with just an instruction set listing. Why are soundcards cheap? Because inside them is a DSP with a native word size of 24 bits and which can only read and manipulate data in that one size, it also has weird instructions that optimize reading an array in an unusual order common in DSPs, not a full-blown ARM core that's made with many more millions of transistors providing features a soundcard will never use. Similarly, the average router until recently had a MIPS CPU with no floating point unit, floats aren't relevant for router tasks. At this point the whole range of programming languages is required somewhere. Straight up machine code, compiled languages, and scripting languages. Focus on theory, general experience, and topics related to robowaifu engineering. Whatever you know will likely be of use somewhere.
>>8608 >At this point the whole range of programming languages is required somewhere. This Anon gets it. We're going to be having all sorts of hardware too. Probably broadly divisible into mobile (ie, onboard the robowaifu) and immobile. >mobile: At least three types and probably more: -edge sensors/actuators. Neuromorphics wants to both combine these together and push large numbers of them out to the edges of the robowaifu. -microcontrollers. Shepherds after a fashion directing all those edge devices, and reporting up and down the chain with the SBCs. -sbcs. The mobile 'brains' that will manage most of the heuristic-type work for the systems. Ofc, they will stay in touch at least intermittently with the immobile resources. >immobile: At least 3 types of systems, probably more: -servers for data, mostly big storage devices. -servers for AI training, lots of GPU horsepower. -gateway systems for securing and defending the rest. On top of that is typical networking, etc. already common in anon's homes. >tl;dr We'll all need a lot of different types of software in the end. So, just make a choice and dive in.
Open file (566.46 KB 1912x1186 graphit-perf_comparison.png)
I'm probably going to look into this at some time: GraphIt - a domain specified language for graphs. https://graphit-lang.org/ https://youtu.be/ptIVf-YlkhY It outputs C++, but optimizes the code based on search algorithms for graphs. Or something like that, lol.
My two cents being an delivering Waifu Engine. Which the renderer is built with C# using unity, then the AI core is built with Python. With Python learn the basics, then learn about what duck typing is vs every other language type system. Then learn how to work with a team by building a domain language around how you communicate. Worst case scenario you end up with a job, but the best thing about Python is there are many resources if you get stuck. A lot of languages like CPP and Lisp or Haskell there are few resources. I know this because I came from a Haskell background, and used it to parallize data workflows using DAGs. You want the language that will have the lowest barrier of entry, any other language will discourage you as a learner. Theres a lot to learn in programming, though the mental models transfer over to other languages, you just need to learn the details and nuances.
>>10495 Thanks very much Anon, I'll look into it! >>10497 >You want the language that will have the lowest barrier of entry, any other language will discourage you as a learner. Theres a lot to learn in programming, though the mental models transfer over to other languages, you just need to learn the details and nuances. I absolutely agree with this notion for a true beginner. But compiled languages like C & C++ bring a critical robowaifu engineering requirement to the table. Namely, efficiency & performance. Because these two compiled languages output machine code that is actually quite close to mirroring the actual underlying hardware (at least when well-written), they don't unduly introduce artificial energy-consumption and wall-clock hits. And particularly for physical robowaifus where Energy is everything, keeping battery drain low is absolutely vital. C & C++ both handle this like a boss.
Common Lisp would be the best option, however you'll need at least some level of proficiency with it to actually use it effectively, which is why something simpler would be better.
Fast.ai will use Swift for Tensorflow in a few years, and already uses it in some paces under the hood. For now they advice using fast.ai and Pytorch to beginners. https://youtu.be/XHyASP49ses >>10553 I have experience in some Lisp, and will go on using it and learn more dialects. However, it always depends on the job, because of libraries and such.
>>128 Probably a more advanced set of programming than what we already have with c++ and machine code with manufacturing machines to make the waifu have perception and be able to not try to punch a hole in a wall trying to reach for the remote.
>>12329 So, have any productive suggestions Anon, or just plying dystopic imagery here?
>>12335 You clearly don’t anon
>>12329 this isn't resolved via language but mechanical strength stepping via complex algorithm tl;dr there's a reason we as biological organisms have adrenaline and super strength when we are under its effects. In a clam relaxed state we are "Weaker" but this is why we don't constantly injure ourselves or break tools and utensils. This is also why we have superior fine motor coordination and dexterity relative to other primates (who have superior strength and power). Our R/Ws are going to need to have different types of actuators for soft touch versus when power is needed (to lift, walk, jump, etc). Otherwise a missed line of code would result in a hole in the wall or worse a sensitive part getting pinched or ending up seriously injured. Simply put, when a R/W is in intimate mode, turn off hydraulic/strong pneumatic actuators and only run e/m servos, solenoids (with resisting springs, counterweights, etc) and nitrile fiber. When R/W has the "all clear" to do outdoor type activity, resume full power. Just one example. Anyway probably belongs in the actuator thread >>12131 or another thread but I've been sitting on this concept for a while and wanted to type it out before it's lost.
>>12361 It's a good point that we'll need multi-level forces within our robowaifus for them to work properly for us Anon
A good, actually very good alternative would be Forth https://en.wikipedia.org/wiki/Forth_(programming_language) Forth was designed when computers had very little memory or power. It's very concise and is productive in an environment where speed and close to the hardware work is needed. It's very small and fast. One of the big benefits is it is made up of words that can be combined. So once you write a routine you can use the word to preform and action. It's used for all kinds of task that demand productivity, speed and are based on limited one of kind type programming jobs or very specific hardware. Look it up I think you will be impressed. There is also a version for ESP32 microcontrollers which I think look to be some of the most powerful best cost versatile microcontrollers built today.
>>12859 Interesting. Is it low-level enough though? We'd need direct access to hardware resources via pointers to do many to the practical things we'd need to to build software for a robowaifu. I know C & C++ offer this, and I'm pretty sure that Ada and D do as well. There are probably others, but tbh there are pretty solid reasons that C & C++ absolutely dominate the systems programming world. And it's not b/c politics or power really, but that other p-word; Pointers.
>>12951 >Interesting. Is it low-level enough though? Yes. Forth used to be, not sure now, THE language that motherboard manufacturers used to boot the computer and set up all the hardware for the OS. It was originally invented by this guy to run a big science telescope. When micro-processors had limited memory and power Forth was used a lot because it's small and very versatile.
>>12951 >Pointers "...A compiled Forth program is a collection of words, each of which contains a statically allocated list of pointers to other words..." "...Forth programmers traditionally value complete understanding and control over the machine and their programming environment. Therefore, what Forth compilers don't do reveals something about the language and its use. Type checking, macro preprocessing, common subexpression elimination, and other traditional compiler services are feasible, but usually not included in Forth compilers. This simplicity allows Forth development systems to be small enough to fit in the on-chip ROM of an 8-bit microcontroller. On the other hand, Forth's extensibility allows "full-featured" systems to consume over 100K bytes and provide comprehensive window-based programming environments. Forth also allows (and often encourages) programmers to completely understand the entire compiler and run-time system. Forth supports extremely flexible and productive application development while making ultimate control of both the language and hardware easily attainable. ..." http://users.ece.cmu.edu/~koopman/forth/hopl.html Since most waifus will revolve around passing messages and reading positions you don't need any large OS. So forth is likely good for waifus.
>>12981 Wow sounds like a pretty compelling position Anon. I'll plan to make time at some point to have a go with it. Thanks.
Here's a link to the history of forth and should give a better idea of it's strengths and weaknesses. https://www.forth.com/resources/forth-programming-language/
One more forth link. It's microForth which is written in C so it's portable. If you search for microForth you get manuals. https://github.com/Earlz/microforth This is just one there ar emany Forths because it's so small people write their own.
>>12986 Thanks Anon I'll look into it.
>>13025 To be fair I need to point out some disadvantages. The way it's constructed you have to think about what you are doing. Modern languages have a ton of programs written already so a lot of it is just stringing together already written code. Forth you will write a lot of it yourself. Since it's a little different it might be harder to read and figure what's going on. My perception is that Forth is kind of like LISP, not that it's any way LISP but since it's a little different and takes some thought to get things going people have a harder time with it. Meaning that it doesn;t grow. It's the same as comparing C++, C, Python and then HTML you have less people able to make good use of these in turn because the level of thought is a little higher in each case with HTML being much easier to grasp than C++. To end on a high note building waifus is not likely to have a lot of ready to burn code for it so Forth is a fast way to prototype and build working devices. The other languages that are easy to use like Python and Java will be too slow and C is too cryptic and liable to errors frequently.
>>13027 "Meaning that it doesn;t grow." Strike that out. I have no idea even what I was trying to say there but that's not right. It happens sometimes you start a sentence thinking about one thing, stop then when you continue you don't tie it together then miss correcting it when proof reading.
>>13028 Heh, no worries Anon it's fine. We all do that. :^)

Report/Delete/Moderation Forms
Delete
Report

no cookies?