5/5 (1)

Three much talked about languages in data science are Python R and Julia. Python, created by the Dutch Guido van Rossum and named after the famous show, is tough to beat and has seen many challengers during its time on top. R, the user friendly, slower and perhaps more sophisticated data science language also won’t be gone tomorrow. Julia, the new kid in town with the odd name. What is it, what does it do well and how does it do that well? These questions I will tackled in an abstract way. The language is focused on data science. Which raises the question: will it beat the other data science languages.

Code readability spectrum

Before exploring Julia, we take a quick step back and look at the programming language dimensions. More specifically, how close a programming language is to machine code. Machine code is ultimately just ones and zeros. If we assume our machine uses the von Neumann structure. This could change when Neuromorphic computing takes off, about which you can read more in the following blog written Wesley Kruijthof: https://digitalstrategy.rsm.nl//2020/09/29/biomimicry-from-neural-networks-to-neural-architecture/. Anyway, in the world of ones and zeros, there are languages which lie closer to machine code and languages which are closer to human readable code. Broadly spoken, there are three categorizations on this readability spectrum: assembly, compiled and interpreted. Assembly languages are designed for a kind of processor, meaning they are highly adapted to the machine they’re supposed to run on. Some families of processors are x86, x64 (the 64 bit version of x86), ARM and MIPS. These families are instruction sets on which languages written to create the code more efficiently. Small sidenote, Apple’s mac’s have long used intel’s x86 processors but are switching to ARM, and will create these themselves. When we go a level higher we’ll see the compiler languages. C, C++, GO, Rust and many more. These programs, when run, are compiled, turned into machine code by the assembler and run directly on the machine. Interpreted languages find a different path to the machine. Interpreted languages are easier for beginners. I’m personally comfortable when writing in Python, R or VBA (Excel specifically). Other interpreted languages examples are Javascript, BASIC, PERL and  Ruby. When these languages are executed, the code gets interpreted line by line. The interpreter reads each line and executes it immediately. For Python the process looks a little bit as the following: CPython translates Python to a C-file. This file gets compiled by C’s compiler GCC (GNU Compiler collection), translated to machine code, which is not executed. This machine code is the actual interpreter and needs Python source code, which is ultimately executed line by line (GeeksforGeeks, 2020).

So where is Julia?

What does this mean for Julia? Or R and python for that sake? Well the interpreter is much slower, though many libraries exist to speed up the process. Both R and Python are interpreted and this is exactly where the difference with Julia lies. Julia looks like an interpreted language, swims like an interpreted language and quacks like a interpreted language, but it is a compiled one. But how? Well compiled code is checked for errors while compiling, which is called static typing. In dynamic typing the checking is done per line at run time. To be clear, the difference in typing has nothing to do with pressing the keyboard. In Julia, Just-In-Time (JIT) compiling is used (Hall, sd). Opposed to compilation, JIT doesn’t compile the entire code in one go, but does it on the fly and opposed to interpreted coding, it doesn’t interpret. But there are already many interpreted languages adjusted to use the JIT way and these languages have existed for far longer. The difference is that Julia’s got some other tricks up its sleeve. Type stability is the notion that only one type can be the output of a method (UCIDataScienceInitiative, sd). Julia scans the code, finds which type is expected and compiles code for that type (Julia, sd). If it can derive which type the output shall be, Julia achieves speeds equal to C.

Does this mean it’s better?

Well not quite. Though speed is a very important, there are many different factors which define a language as better or best, which obviously aren’t defined terms here. R for example is rather slow. It’s not that some guy from the R-team, while years into development, in a conference said: ah sh*t, we forgot to implement speed. R is for data scientists and is made simple/convenient. It’s also not made to process terabytes or produce time critical results. A complete beginner will get used to R quicker than to Julia and the same applies to Python.

Versatility is also important. When you find your way into a job application for data scientist and are asked why R is popular in the field, the answer shouldn’t be that it’s easy to learn. R’s CRAN has over ten thousand official packages which are solely made for data analysis and related fields. Dwarfing Python and demolishing Julia. The amount of mature packages is directly related to the community of a language. When more people use a language, it gets streamlined. Improvements will be made when many people spend a lot of time. If you google “Biggest R contributor” I can assure you the name Hadley Wickham will pop up very quickly. It’s safely argued that R wouldn’t nearly be as popular without his contributions. Which are both easier to use packages, faster packages and books explaining those. Python on the other hand is the most versatile. Though seemingly contradictory to the dwarfing statement earlier, PyPI has the most packages of the three with quite some distance as well. That’s because Python is also very well developed in other fields like web development, databases and webscraping. (CRAN, sd)

Future

So Julia’s prospects? Julia has got an annual growth rate of 101% (Simplylearn, 2020). Meaning the community is growing and the packages are expanding. If you were to take only one thing from this blog about Julia. It’s that Julia combines the best of two worlds: high performance and ease of use. Somebody who wants to get the best of both worlds, might actually write programs in the two different world. Prototyping in one, and implementing in the other is no longer necessary. Solving this problem will create a user base, which in turn sparks some good old positive network effects. I haven’t even named the possibility to call Python and C packages from Julia yet. Also, Julia might be able to learn from its predecessors. Creating the equivalents of popular packages. Almost forgot: like Python & R, Julia is free (looking at you Matlab and others).

 

I’m no computer scientist and currently just an aspiring data scientist, so I’m happy to hear any additions or errors in my blog. The future of Julia is discussed intensely in the science community, so I’ve heard. What I ‘d love to hear is your opinion on its prospects. Is Julia going to make it to the top or not? And why?

Bibliography

CRAN. (n.d.). Contributed Packages. Retrieved October 8, 2020, from CRAN: https://cran.r-project.org/web/packages/index.html

GeeksforGeeks. (2020, June 8). Compiler vs Interpreter. Retrieved October 8, 2020, from GeeksforGeeks: https://www.geeksforgeeks.org/compiler-vs-interpreter-2/

Hall, M. (n.d.). Julia in a nutshell. Retrieved October 8, 2020, from AgileScientific: https://agilescientific.com/blog/2014/9/4/julia-in-a-nutshell.html

Julia. (n.d.). Performance tips. Retrieved October 8, 2020, from Julia: https://docs.julialang.org/en/v1/manual/performance-tips/

Simplylearn. (2020, January 21). Things to Know About Julia Programming Language. Retrieved from Simplylearn: https://www.simplilearn.com/things-to-know-about-julia-programming-language-article#:~:text=It%20may%20seem%20like%20an,annual%20rate%20of%20101%20percent.

UCIDataScienceInitiative. (n.d.). Why Does Julia Work So Well? Retrieved October 8, 2020, from UCIDataScienceInitiative: https://ucidatascienceinitiative.github.io/IntroToJulia/Html/WhyJulia

No ratings yet.

Upcoming November, the tenth, Microsoft will release the Xbox series X and S. However, Microsoft doesn’t hide the fact that their image of gaming isn’t centered around the console. Microsoft’s Game Pass is rising quickly and it seems to be no secret that Game Pass will get prioritized over the traditionally expensive console.
Microsoft consoles will go head to head with Sony’s PlayStation 5 into the console battle. Quite accurately explained, the console generation battles go a little something like this: Sony and Microsoft introduce a new console once every 5-7 years. Both machines are heavily improved on technical features, allowing better gaming experience. Both claim their machine is ground breaking and next-gen, completely disregarding the superior technical possibilities of PC gaming. And in the end, Sony wins (VGChartz, 2020). As a PlayStation owner of all generations, my opinion can certainly be taken with a grain of salt, but Sony did win in actual console sales quantity.
The strategies for the next battle, of the consoles released this year, show obvious comparisons with the last battle. In 2013 Sony’s PlayStation 4 won with a straightforward tactic; console gamers want a console, let’s give them one. It improved on several aspects but the idea is clear. Microsoft on the other hand viewed the console as a home video hub. It also bundled with a Kinect motion system, making the device expensive. About a 100euro more expensive than the PS4. Making the expensive and slightly inferior machine significantly less popular. The resemblance with the current battle lies in the fact that Sony once again recycles the straightforward approach while Microsoft is trying to disrupt the whole industry by offering subscription based games. Every game related thing Microsoft does seems revolved around the game pass. Phil Spencer, executive at Microsoft, spoke the following about it: “If this is not the year when a family wants to make a decision to buy a new Xbox, that’s OK. Our strategy does not revolve around how many Xboxes I sell this year. We’re focused on delivering services through Xbox Game Pass, which allows people to build their library for a monthly fee.” Microsoft is trying to get as many people as possible on a subscription base and has several technique to do so (Gamesindustry.biz, 2020). Starting with the unprecedented Xbox all access scheme, in which the consumer gets the console and game pass for a subscription for two years (Xbox, 2020). Then it is nice to know that game pass allows gamers to access their game library not just on the next gen console, but also on PC and current console (Xbox, 2020). Furthermore, the cloud solution xCloud is part of the game pass subscription, meaning that if cloud based gaming takes off, there is no reason to buy a console. Finishing off with the pricing; series S is only half the price of the series X and certainly lower than the PS5.
All the things named above clearly indicate that console revenue isn’t the priority. However, Microsoft amassed a huge fanbase of the Xbox, which can’t just be let down. The subscription based library of games, sometimes even Xbox exclusive, is playable on for high performance PC and the current generation console works fine still. So it seems Microsoft is trying everything in it’s power to disrupt gaming as it is now, while taking down its own console. Battle after battle was lost, but it might just be that Microsoft will win the war. The results are rolling in, starting upcoming November when the consoles release. A month which will be essential for Sony. But Microsoft releases just another machine to support the game pass. A machine which is brought to life, not for income but for the fanbase. A next generation, cutting edge technology, high priced product which even Microsoft finds unnecessary. Or in other words, just a legacy product.