Jupyter, Mathematica, and the Future of the Research Paper
paulromer.net
paulromer.net
The tie-breaker is financial. Jupyter is winning because it's free, not because it's social. It becomes social because of widespread adoption, and it get's widespread adoption because it's free.
I love Jupyter, love love love. But there's a lot of hyperbole and opinion here. Mathematica is just a for-profit business, it's that simple. And it wouldn't be fair to deny the example that Mathematica, Maple & Matlab have set for free solutions like Jupyter.
There's nothing dishonest about a for-profit business. And if Mathematica wants to keep PDF export for themselves, so be it, that's their right. What's dishonest is expecting technical software and service for free and calling people names like vandals if you don't get it. Just celebrate Jupyter and enjoy that people are doing great work you get to use without paying. I don't love Mathematica or it's founder, but there's no real need to impugn Mathematica in order to make this point.
Well for you maybe – and we strongly believe it should be – but it's built on top of thousand of volunteers hours, grants money (Thanks Sloan, Helmsley, Moore) and donation from companies (Anaconda, microsoft...) and individual, and partners. NumFOCUS (https://www.numfocus.org/) manage all of that it's a 501c3 tax deductible ! If Jupyter is of help to you (or your company, organisation) think about contributing back (Code, Dev Time, Design, UX, translation, Legal, ...)
Much Love from the Jupyter Team.
We are also striving to make contributing to Jupyter / open source recognized in Academia - for people like Fernando to get proper recognition, and future generations of scientists to have incentives to do this kind of work.
And thank you for this! Do you have a post about what efforts are going on? I myself am trying to find out how to navigate academic spaces while spending significant time doing open source development. I am curious how you're pushing for changed incentives to help people in this kind of position out.
Understanding what it took for Jupyter to be where it is, and how much work it took, who pushed for it, and how many people had to work to make it happen, this is important.
You're absolutely right to emphasize that it's not free to build. I worry that promoting a mentality like the author's that open source and free to use is honest, and for-profit is "dishonest" dramatically undervalues and undermines your work as well as the funding model. If we treat Jupyter like it's our right, and not a miracle of funding and hard work, we will make it more difficult for Jupyter to attract help and to succeed.
Same goes for the rest of the data stack: Python itself, Numpy, Scipy, Scikit-Learn, Pandas, etc.
Jupyter is great, it's revolutionized online education for example. But it's not as full featured as Mathematica, and like most Foss projects, slower to fix bugs, aka you get what you pay for. Both have a place unless you have hired, 100% dedicated help on juptyer. This is much like public radio vs corporate. Together they do the market needs.
Jupyter is not GPL but MIT, and volunteering is rarely about credit.
> The tie-breaker is financial. Jupyter is winning because it's free [...]
It is a bit more nuanced than that. Personally I do not pay from Mathematica usage, so why do I like Jupyter more?
The Mathematica notebook interface is horrible. You may go "oh, neat" the first few times you try it then, at least I, get more and more frustrated on all the idiotic issues
* Indentation/text-wrapping. Write a long line that starts wrapping, it gives it a little indentation to signify this, an your next line that you have indented is slightly more indented, but it is really hard to see, so you have no idea of where your line-breaks are.
* Brackets. "[" are used for function calls and for part-specification. The number of square brackets in your expression makes it necessary hard to read when it is big enough.
* Jumping text. The notebook interface does not have the "auto-complete brackets" (maybe v11 does), so you add your first, all the text in the Cell gets reformatted and you have to find the fucking place you wanted end the bracket. This is akin to working with images in MS Word.
* Exporting. The notation is just ugly, fine, that is personal, but "Sin[]" as "sin()"... ok. Ah, good, it has a "Copy as Latex", nice... "Sin[]" -> "\text{Sin}[]". Really? Who in the world uses "\text{Sin}" for the sine-function and square brackets for function-calls when typesetting maths in Latex?
It is just a complete nightmare to try and incorporate these things into your workflow, at least for me.
Jupyter just behaves as you'd expect. Just that it is so much smoother to work with wins. For me, I do symbolic calculations, SymPy can do some things much easier than Mathematica, but a lot of things it can't or you have to work some more to get going. That Jupyter allows you not to have an aneurysm every day at work, which makes you actually wanna spend the extra time working it out.
You complain about Sin[] but not about ugly things like np.sin()? Also I think you're looking for
Sin[θ] //TraditionalForm //TeXForm> Sin[θ] //TraditionalForm //TeXForm
Did not know of this. Thanks!
So TeXForm[Sin[x]] gives "\sin (x)"
And while I swear I've had rotorblade's problem before, right now "Copy as LaTeX" seems to do this too.
There is a very easy to implement modification you can make to allow the keyboard shortcuts `CMD`-[ and `CMD`-] for the double bracket symbol, which is described here:
http://szhorvat.net/pelican/pages/mathematica.html
I rather wish Wolfram would provide these keyboard shortcuts in the base software
This is it really
Back in the 90s I was using a program called MathCAD, it provided a “notebook” interface by running as a plugin to Word 6. In terms of general usability and experience, 20-years-ago blows away modern-day Jupyter and it’s silly “cells” interface, which it does not because it’s better but because it’s trying to force itself into a web browser. I haven’t used MathCAD since but I bet in 2018 it’s amazing.
I think few people who have used the commercial tools think Jupyter is better. But the commercial tools are soooooooo expensive...
We acknowledge there's a lot to improve in Jupyter, and some discussions in this post make excellent points (many of which we'd like to make progress on in the future). But the Jupyter team did probably use most/all of the modern scientific computing platforms, MathCAD included (and Maple, Mathematica, IDL, Matlab, Gnuplot, ...) at some point in our careers. We typically make our choices with reasonably good knowledge of the landscape.
We make mistakes, or our tradeoffs may be different than the optimal ones for your use case. But lack of knowledge of these tools is rarely the reason :)
I'm waiting impatiently for the coming revolution of non-web collaborative internet-connected rich client applications.
Of course, it's not web-based and it requires some basic knowledge of Emacs, but functionality-wise is so much better that it ends up being frustrating to use Jupyter when collaborating with other people. There are some workarounds (EIN, and ob-ipython modes help, but it's not quite the same)
This feels like the argument against Slack in favor of IRC. I don't doubt it's true for you that org-babel is better, but there isn't really even that much comparison between Jupyter and org-babel, they are very different things.
One of the more compelling reasons to use Jupyter is inline images and plots. Another one of the more compelling reasons to use it is it basically comes bundled with everything you need, python with it's own virtual environment that doesn't modify your system python, and a bunch of amazing libraries, numpy, scipy, matplotlib. Not needing to know emacs helps too, but that's really low on the list.
> Jupyter uses a rather obtuse js on format which is not practical to work on directly.
This is a wierd thing to say. Jupyter has download as .py right in the file menu. Not to mention download as Markdown, HTML, LaTeX or PDF, if you want to present or share instead of export the python.
You can do those in org, since forever. I meant that the files themselves are plain text, which is good for sharing, version control, etc. ipynb files embed images with Base64 encoding, which makes seeing diffs a major PITA.
>This is a wierd thing to say. Jupyter has download as .py right in the file menu. Not to mention download as Markdown, HTML, LaTeX or PDF, if you want to present or share instead of export the python.
You can export org to those formats and more, but the main part is that you can organize your code (e.g, tangling to different files, etc.), and use other languages than python (or whatever other kernel you're running in Jupyter).
Plus, you get to use a proper text editor instead of editing on a browser textarea.
It sounds like org mode does not store images inline, since you don't like the b64 images in Jupyter? I imagine that could go both ways depending on what you want. Even for diffing, you might like to see that the image changed (though I generally agree that diffing images is not ideal). If you have to package images separately when sharing, then wouldn't inline images score a point for easier sharing?
> use other languages that python
Agreed, and that is powerful, but this also goes to my point. Jupyter is made for python and org mode isn't. Jupyter comes with python and libraries, where with Emacs you're on your own to figure out how to install and use different languages, and setup your environment so Emacs can see them. You're on your own to bring in images and plots, where Jupyter already understands commands that output images.
> Plus, you get to use a proper text editor instead of editing on a browser textarea.
Yes, but the editor is Emacs. :) For a lot of people the browser textarea is a bonus because it's the same thing you use everywhere. It's uncomplicated and non-technical and utterly consistent, even if lacking any power.
I guess that depends on who you're sharing with and how. If you're using git, you can just add them to your repo (only they won't clobber your diffs). If you're just publishing, you can export to HTML and put it somewhere online.
>Agreed, and that is powerful, but this also goes to my point. Jupyter is made for python and org mode isn't. Jupyter comes with python and libraries, where with Emacs you're on your own to figure out how to install and use different languages, and setup your environment so Emacs can see them
Well, that requires a minimum of computer literacy, but rocket science, it ain't. I agree that Emacs is not for everyone, but if you're doing data analysis (and presumably something with the results), it's certainly not out of reach, and good tooling is worth learning.
I agree that Jupyter is "easier", but it's not simpler. And also, it's limiting in what it can do. So yes, you might get up and running a bit faster, but you're setting yourself for frustration down the line.
> Yes, but the editor is Emacs.
Precisely, it's vastly better than the alternative. If you don't like the bindings, you can use Evil (this is my choice, despite never having been a heavy vi user), or CUA mode, and get a more familiar experience for people used to Windows.
Adding git as a dependency to sharing is too much in general, that would prevent sharing.
I'm fully aware of that, my point was that functionality- and ergonomics-wise, you're better off using Emacs/org. I agree that it's not for everyone, but for people working on data science, it's a much more powerful tool.
>Adding git as a dependency to sharing is too much in general, that would prevent sharing.
I think we probably have vastly different target audiences in mind.
I have a shortcut that lets me edit any browser text area in MacVim, which works out really well in the Jupyter notebook.
What I wonder is why are Jupyter notebooks so popular? I've tried them, but perhaps I don't understand the workflow in the way people use them. Are they intended to be works in progress where one explores ideas and looks at results? This is how I've usually used them and Mathematica's notebooks as well.
For me though, my Jupyter or Mathematica notebooks end up with lots of false steps and dead-ends. To prepare a presentable form of the results I have to go back and edit the notebook to such a degree that it doesn't feel like working in a notebook as much as it is working in a complex document with a really weak set of editing tools. To be fair, I haven't used Jupyter or Mathematica for a couple of years and things may have changed. Maybe HN readers can help be understand how to more effectively use these tools.
How easy is it to share a Jupyter notebook with someone else, versus sharing an org-mode notebook? Think about the recipients rather than how hard it is for you. Assume the person you share with needs to be able to re-run your notebook. Assume the person you share with isn't using the same operating system that you are. Be honest and think about all the steps involved.
With Jupyter, I send a single install link to Anaconda. With Emacs + org mode + python + numpy + scipy + matplotlib, I might not even know where to begin if the recipient is using Windows or Mac. The recipient needs to install emacs, and they need to know how to use Emacs. They have to know how to install python packages (and possibly a package manager), and ideally be able to use virtualenv too.
If I need different libraries than what Anaconda comes with, I can put the commands in the notebook, and I won't mess up the recipient's system version of Python. With org-mode, they'd need to figure it out on their own.
If I choose to use a different language than Python in my org-node notebook, the recipient will have to install that programming language and environment themselves before they can use your notebook.
It's the simplicity and bundling of dependencies that make Jupyter popular. All the flexibility and power you get with org mode comes at a high price.
If they want to actually work on the notebook and keep their changes, they'll probably need to install it (or switch to a platform like Cocalc). But Binder is enough to let them view, run it and make experimental changes, without installing anything or creating any new account.
Well, now you're talking about a whole environment. Use docker, or VMs or whatever. There's plenty of options for doing this (we actually use Docker for Jupyter, to make sure we work off a consistent install, so I guess sharing can be non-trivial if you walk a bit off the beaten path, anyway)
> If I choose to use a different language than Python in my org-node notebook, the recipient will have to install that programming language and environment themselves before they can use your notebook.
Sure, and if you need to use another language from your Jupyter notebook, you can't. I don't get your point. Yes, if you need to use a tool, you need to install it. It requires a minimum of computer literacy.
> I don’t get your point.
My point was Jupyter inflicts far less technical burden and debt on other people. There’s a huge downside to custom environments. If that’s not obvious, that explains why it’s hard to understand jupyter’s popularity.
> it requires a minimum of computer literacy
I’m sure you didn’t mean it that way, but this borders on white tower programmer centric dismissiveness.
Personally I know how to install all the things an org mode notebook might require, but I don’t want to. It takes too much of my time and energy. Expecting other people to adopt your choice of tools and environment by arguing that your tools are basic literacy is presumptuous. Use org mode, I’m wildly in favor of people using tools that they like, and you like it. Just don’t claim it’s objectively better on all axes when it’s not.
You're again comparing apples to oranges. If you want to use Jupyter, you need to install a whole lot of stuff (easier to do if you're using Anaconda, agreed). It's "not simple" enough that there are public docker images for it. My point is that when tooling is an issue, you can do the exact same thing with an org-based setup.
>My point was Jupyter inflicts far less technical burden and debt on other people. There’s a huge downside to custom environments. If that’s not obvious, that explains why it’s hard to understand jupyter’s popularity.
I agree that there's difficulty in achieving reproducible environments, and that's why there's tools that handle that for you. I think that much of Jupyter's popularity comes from the fact that it's easily available, and based on Python, which let's non-programmers get started moderately quickly. I just don't think that it's the best choice. This is not to shit on Jupyter, I've mentioned a bunch of advantages of using org (which by the way, you haven't really addressed, other than focusing on the setup aspects)
> I’m sure you didn’t mean it that way, but this borders on white tower programmer centric dismissiveness.
It's not, really. My point is that when you're doing presumably advanced analysis on presumably complex data, you (should!) certainly have the capacity to deal with tooling.
Not entirely true, BTW.
https://blog.dominodatalab.com/lesser-known-ways-of-using-no...
Seen that way, sharability is just a nice bonus. Yeah, I'd never try to turn anything directly from a jupyter notebook into a serious way of sharing research results---but for a quick and dirty "here's what I did" on a blog or email or something, or to show students, it does the job.
Kind of odd, but repl on steroids is really just how dim the modern view of the old lisp machines is. Most of our neat repl tricks were just taken for granted on some of those machines.
That being said, it's still pretty sweet!
[1] https://github.com/syl20bnr/spacemacs/tree/master/layers/%2B...
I often wonder the same. I'm now kinda forced to use them, since they're the tool of choice in my team, but it's a major pain. I think that they're "easy" (in the way Rich Hickey describes "easy" vs "simple") to get started with (just as python, btw), but their lack of flexibility comes back to bite you sooner or later.
For me, I've resorted to using a mix of EIN and ob-ipython, and then copy/pasting as needed to keep the stuff in synch with Jupyter. I'm now considering writing a simple ox-jupyter exporter, and maybe some import functionality, to avoid dealing with Jupyter directly.
John Kitchin has written a basic exporter (http://kitchingroup.cheme.cmu.edu/blog/2017/01/21/Exporting-...), but it doesn't support inline images (it should be simple to fix this), and it doesn't really adhere to the ox- style of exports, so I think I'll probably do my own instead of adapting his (it'll be a good learning experience, for sure).
You can do this with RMarkdown as well, despite the name, and run notebooks in RStudio. Much better experience than Jupyter in a browser.
Except engine='R' (default), all chunks are executed in separate sessions, so the variables cannot be directly shared.
1. Plain text format, git and git diffs work
2. You can combine many languages in a single document, and every code block can be part of a separate session, as an analogy to Jupyter Notebooks, you can have multiple kernels backing a single notebook and you can decide what kernel you want the current code block to run in.
3. You can edit a code block in the major mode for that language, i.e. you get all the features of Emacs while editing code: documentation, auto-complete, snippets and anything Emacs can do, and Emacs can do a lot :)
4. You can have internal and external links to any part of the document (or any other org-mode file) within the editor which get exported as links in the HTML file too. Want to refer to a code block you used before, just name it and drop a link. Extremely useful in binding the whole document together.
5. Literate Programming support -- You can decide the order the concepts are introduced in according to the human reader, not according to the execution order the machine demands it to be in:
#+NAME: named_code_block :eval no
function_not_defined_yet()
#NAME: complete_code_block
def function_not_defined_yet():
print("nice function innit?")
<<named_code_block>>
6. With the internal and external links and <<named_code_block>> (NOWEB syntax) the org-mode file is closer to being a hypertext file than Jupyter Notebook even though Jupyter Notebook is the one running in a browser.
I have covered only the major features of org-babel, I haven't even covered all the features. I love Jupyter Notebooks too, but org-babel is something else. I am currently working on a toy ray tracer in Clojure in literate programming style and loving every moment :)
Just to give an example of what can be done using org-mode, this is the project I am using to grok Literate Programming -- http://ray_tracer.surge.sh/
The whole thing is generated using one org-mode file -- https://gitlab.com/snippets/1710454
and this org-mode file is the one I work in, it will eventually generate the source code in separate files too, once I have finished the project.
This is what my setup looks like -- https://i.imgur.com/mqi8vPR.png
Anyways I can't convince you to use it, but hopefully I can convince you to give it a try, it isn't easy but it is worth it :)
I'd add that you can also benefit from other aspects of org, such as project management functionality, outline editing, table editing, tables with formulas, direct git integration via Magit, etc., etc.
I agree that getting people on Emacs is a non-minor issue, but using something like Spacemacs with CUA-mode enabled could go a long way towards acceptance.
I am really interested in the specific advantages MathCAD had/has, and the problems you see with cells and being in a web browser.
> the coming revolution of non-web collaborative internet-connected rich client applications.
why? how is the web not up to this?
My end goal is a document, and for that I either want a programmer-grade text editor, or a full-featured word processor. Filling in a web form is a terrible way to write a document. My tool of choice at the moment is RStudio - all the interactivity you could possibly ask for, and very easy to edit complex files then render them to a great-looking PDF with the output of my code interspersed with paragraphs of text, footnotes and TOC and all the rest.
MathCAD was later bought by PTC, and further went to shit. It amuses me how some companies can self destruct.
That would be a couple of years after I last used it. What a shame, at least I have those memories. Really loved how low-friction working with it was, and how trivial it was to generate high quality documents or reports, there was no "write it up" stage, you were done basically as soon as you had done the actual work. Then just print and go! Still can't really do that with Jupyter.
Started using the DOS version in 1990 in college.
Bought version 3.1 in ~1992, Windows version running it under OS/2. Could afford to buy it then, even as a student. Ah, OS/2 was also good.
It was < $100 for each version upgrade until 2000, when they switched to maintenance based licensing, soon bumping the price to $495.
I had a home license which I dropped. I still had a work license, but then some sales twerp said I was using a pirated license at work, and wanted to speak to my IT department. This is a >100k employee company, so serious consequences for software piracy. I faxed them my valid license certificate, and told them to stuff it, dropping my renewal, and buying Mathematica.
Now I see it's Mathcad Prime, a subscription based license, for $800 year. They were purchsed by PTC several years ago, which only made things worse.
This is textbook case of screwing over your user base, cha$ing corporate accounts. I don't know anyone who uses it now. Many people used it back in the day.
Sad thing is I don't even use Mathematica much. Too painfull most of the time. I don't like the interface, but it's better than browser based Jupyter. Just forget trying to install Jupyter and dependancies on a corporate managed Windows computer. I just use papaer and calculator. I swear things have regressed from 20 years ago. At least I can still buy a new RPN calculator.
Looks like someone has: http://comppad.sourceforge.net/
It can be both financial and social: individual institutes are all too happy to pay for Mathematica licenses, and consequently institute members can use it to produce reproducible research with it. However, the reproducibility of the resulting notebooks is drastically hindered by the fact that a reader essentially also needs to pay for Mathematica to get the full benefit out of these notebooks (even if they are readable without Mathematica). As a consequence, few people bother using it even though they can afford to. Social drivers disincentivise its usage.
By a similar dynamic, Git beat out the competing DVCS: in this case mostly technical rather than financial factors that drove individual actors to prefer Git over alternatives (due, to a large part, to GitHub). But many people don’t actually care about technical considerations (or even prefer other systems over Git in this regard). What people most care about is seamless integration. In the end, a social driver caused Git’s adoption.
That said, I'm still missing a free but powerful tool for symbolic computations like Mathematica or Maple.
> The tie-breaker is financial.
Exactly, it cannot be emphasized enough. Of course as a student you get these powertools for a small price or even for free. But if you are not in University, those tools are super expensive. For a reason but there is still a need for far more open source in this area.
EDIT: I'm just realizing there is Sympy, niiceee...
I am personally more a fan of Julia than Python, and Julia + Jupyter is an awesome combination.
Some observations I have are that they're both great. Python is a nice open source scripting language, but getting libraries to work can sometimes be a pain. Mathematica is basically install this and everything is included. The Mathematica documentation is amazing and it is really simple how to do most things. The whole iPhone "there is an app for that" is equivalent to "there is a function for that".
Graph Theory works flawlessly in Mathematica. In Python, there is a module to Graphviz. Let me know if Python has something new though. There are a lot of other examples. Mathematica's Import[] function can read over 150 different file types including: CSV,.XLS, genetic encoding files, optimization files....whatever. It is usually far easier and more consistent than finding a corresponding Python library and struggling with the install and minimal documentation. Let me be clear that Python is awesome and rocks and i think Jupyter is moving it in the right direction. I just feel like many dismiss Mathematica as something that does Calculus homework rather than what it is today which is a massive 20 million LOC conglomeration of C & Java & Wolfram language that does everything from Statistics, Machine Learning, Visualization, BlockChain, 3D printing, NodeGraphs, data sets and analysis...etc in a single consistent package. It is expensive and proprietary and certainly has its own faults, but a lot of that cash is funneled back into a great product.
The problem with Mathematica from a science point of view is that, being closed source, means you can't independently ensure the calculations are happening correctly. To be replicable, science involving data needs to use open source tools.
Have there been any high profile failures root-caused to Mathematica (or MATLAB or any similar product) getting its sums wrong? I can find any news stories etc. Plenty of serious calculations were and are done on “closed source” HP and TI calculators too. Every serious scientific instrument with its own data capture uses a binary blob too somewhere, so if that’s a problem for you then you can’t even trust the raw data!
And even if you have all of the code - you still need to worry if the proprietary, closed FPU is working “correctly”.
This sounds very much like a post-hoc justification for “its free as in beer”. Do you think Wikipedia is more trustworthy than real references too? What about blogs?
But consider http://www.bbc.co.uk/news/science-environment-39054778
"Science is facing a "reproducibility crisis" where more than two-thirds of researchers have tried and failed to reproduce another scientist's experiments"
I don't think that can be handwaved away as "OMG closed source software!". Especially since all the scientists in a given field will have access to the same software anyway. Give them open source and the issue will persist, and we both know it because the root cause isn't anything to do with the license of the software
Of just consider peer review. How often does a reviewer today actually review the code and data used in a study? As far as I know this is essentially never done. That would involve seeing the code, running it, maybe messing a bit with it.
Citation very much needed for that. Because you can very easily find that 6 months or a year later you update your dependencies and everything is now broken. I recently came back to a Python project after a year, updated my packages then realized: I simply cannot be bothered to unpick the mess that resulted just to add one trivial feature. Whereas the poster child for backwards compatibility is closed-source and proprietary.
Science is not reproducible because there are no incentives for it to be so, despite everyone paying lip service to it. It's extra work and helps those who are competing with you for grants, after all. That's a social problem, not a software one. The software is irrelevant.
Secondly, of course, open source means that if you're not sure why two versions/functions/libraries are giving you different answers, you can go and find out. I accept that a lot of people may not have time for that, but I don't think you can fix that problem unless it's possible to dig down and follow the working.
Finally, 'reproducible by anyone with a computer' is a lot better than 'reproducible by people who buy a license for the tool I used to do it'.
Better record which compiler you used too, and what flags, and every version of every library and everything else. It’s not as simple as you make out and it’s far from guaranteed that all those packages will still be available or compile on your OS.
Commercial software vendors are usually not keen on you downloading and running a version of their product which was superseded four years ago.
I guess you must not deal with vendors much because generally they are fine with this. It’s part of the support agreement usually, just another service. Getting an “obsolete” version for whatever reason has never been a problem for me.
By “anyone with a computer” you mean “anyone who can exactly reproduce my configuration which I don’t even know myself for certain”
The typical problem paper has a small data set, on which the authors tried 20 different things, one of which achieved p<0.05 and got published. The result tells you nothing meaningful about the world (or more often, tells you something about how its authors wish the world worked). But re-running their code on their data set will not reveal the problem.
Now being open source does not prevent such errors. What I think is far more important here is open issue tracking: companies like Wolfram may not be exactly eager to let you know that there are serious bugs in their products (and what they are) unless they must. Being able to fix the bug yourself can of course be a real perk for some user's, but realistically, it's out if question for most.
I get the feeling that many of the Jupyter fans commenting here have only ever used a Jupyter setup and are unaware of the wider industry
Excel and MATLAB can't be used for real science?
Abstract: "The spreadsheet software Microsoft Excel, when used with default settings, is known to convert gene names to dates and floating-point numbers. A programmatic scan of leading genomics journals reveals that approximately one-fifth of papers with supplementary Excel gene lists contain erroneous gene name conversions."
For a fun look into the high standards of the field, a few years ago Piketty made the mistake of sharing his Excel files, in which all sorts of crucial adjustments were hard-coded into tables of data...
https://marginalrevolution.com/marginalrevolution/2014/05/pi...
One thing in Python's favor though might be depth in certain categories. The machine learning stuff in Mathematica is very nice and high level if you want neural networks, but if you need PSO or GA, you'll probably have to write your own or grab someone else's notebook.
It was difficult for me to support closed-source software as I've always supported linux for this reason.
As far as ensuring accuracy of calculations, having a very large and highly technical user base over several decades helps, but I'm not sure how much this is used in theory. If a statistician publishes a paper using R, is anyone really going to check the R module source code? I bet this is a rare occurrence.
"Membership in an open source community is like membership in the community of science. There is a straightforward process for finding a true answer to any question. People disagree in public conversations. They must explain clearly and listen to those who response with equal clarity. Members of the community pay more attention to those who have been right in the past, and to those who enhance their reputation for integrity by admitting in public when they are wrong. They shun those who mislead. There is no court of final appeal. The only recourse is to the facts.
It’s a messy process but it works, the only one in all of human history that ever has. No other has ever achieved consensus at scale without recourse to coercion.
In science, anyone can experiment. In open source, anyone can access the facts of the code. Linus Torvalds may supervise a hierarchy that decides what goes into the Linux kernel, but anyone can see what’s there. Because the communities of science and open source accept facts as the ultimate source of truth and use the same public system for resolving disagreements about the facts, they foster the same norms of trust grounded in individual integrity."
The entire blog post is worth a read.
Oh please. Dare to ask what is the best of anything and prepare for an epic flame war.
https://twitter.com/ChrisRackauckas/status/98552939470474035...
So your remark is pretty spot on.
1. Mathematica has an easy way of sharing notebook. I just run "deploy" command which turns notebook into publicly accessible webpage, hosted by wolfram, here's an example -- https://www.wolframcloud.com/objects/user-eac9ee2d-7714-42da...
2. Mathematica has more active community. Mathematica-specific questions are likely to be answered within an hour by experts on https://mathematica.stackexchange.com/
3. Mathematica has better tools for simple interactivity. I like to throw in "Manipulate" for a simple graph with a draggable constant, or go to http://demonstrations.wolfram.com/index.php for an idea for more complicated demonstration to use in a presentation
4. Mathematica has more options for advanced visualization, and interfaces are more uniform since graph drawing, 3D drawing, and other kinds of visualizations are developed within a single system. Some examples https://www.wolfram.com/language/11/new-visualization-domain...
1) Binder makes that a git push away https://mybinder.org/ Want to check the discovery of gravitational waves ? Go ahead ! https://github.com/minrk/ligo-binder You know the nice thing ? it does not require you to opt-in, as long as a repo is public you an run it. So you don't need have to deploy, or know it exists. we are _already_ doing that for you.
2) Jupyter is "Just" the frontend. StackOverflow have matpltlolib, numpy, sympy, .. tags. We don't the subdomain (yet), and I actually prefer to have tags to have better searching :-)
3) Sure it's called ipywidgets (https://ipywidgets.readthedocs.io/en/latest/), that's the tech. From ipywidgets import interact, and @interact as decorator on your function... that's it.
4) For convenience Library that use ipywidgets for 3D see https://ipyvolume.readthedocs.io/en/latest/animation.html (Hey it also support VR !) See https://www.youtube.com/watch?v=nZ3HQpSXn2U that will blow your mind.
We'll try to be better at advertising our features !
Check out jakevp's book for example:
https://notebooks.azure.com/jakevdp/libraries/PythonDataScie...
Clone to run.
Also try Jupyter Lab (experimental) - closer to an IDE than plain notebooks. right click on a Library (repo) and select Open in Jupyter Lab.
[disclaimer - our team's offering]
Jupyter encourages individual integrity; Mathematica lets individuals hide behind corporate evasion
I have no idea what he's talking about re PDF export either. I print to PDF all the time, to email people a static document to look at, etc. It works just fine. (Whether you can you make book-quality formatted text easily, I've no idea, never been tempted to try.)
Whenever I've used the "Save As..." command, choosing PDF as the target, I've also only had good quality output.
https://mathematica.stackexchange.com/questions/68893/save-a...
This segment, with it's hyperbole, ridiculous black and white simplification of complex matters, and tone of (over)confidence is unmistakably typical of her writing style:
"Jupyter rewards transparency; Mathematica rationalizes secrecy. Jupyter encourages individual integrity; Mathematica lets individuals hide behind corporate evasion. Jupyter exemplifies the social systems that emerged from the Scientific Revolution and the Enlightenment, systems that make it possible for people to cooperate by committing to objective truth; Mathematica exemplifies the horde of new Vandals whose pursuit of private gain threatens a far greater pubic loss–the collapse of social systems that took centuries to build.
Membership in an open source community is like membership in the community of science. There is a straightforward process for finding a true answer to any question. People disagree in public conversations. They must explain clearly and listen to those who response with equal clarity. Members of the community pay more attention to those who have been right in the past, and to those who enhance their reputation for integrity by admitting in public when they are wrong. They shun those who mislead. There is no court of final appeal. The only recourse is to the facts.
It’s a messy process but it works, the only one in all of human history that ever has. No other has ever achieved consensus at scale without recourse to coercion."
> Now, Jupyter is the unambiguous technical leader.
is pure fantasy, imho. SymPy is still two decades behind Mathematica in large swathes of symbolic computation.
It may be that, for the things that the author wanted to do, the Python libraries were a good fit (it seems he was working on NLP), but overall, I just don't see it.
Jupyter has a lower friction to adoption and usage than Mathematica, for a definition of friction which encompasses ease of acquisition and sharing. I include economic considerations in the ease of acquisition and sharing. Lack of proprietary walled garden lock-in/lock-out factors in as well.
People are also likely considering the longer term scenario, whereby data, model, and information interchange has been hindered by proprietary formats (the "wall" in the walled garden) and lack of complete information on how to get information in and out. Which is what the OP was complaining about, as they were not able to easily construct a publication quality preprint/submission from one, but could do it easily from the other.
Some of these sources of friction are effectively "own-goals", that is, you increase friction in such a way as to prevent something that people need to do, to be effectively impossible. Or you hide it. Or disable certain groups from using that functionality.
Then the question is balancing the longevity of the format, the proprietary value against alternatives. Increasingly, people are less interested in this friction for a number of critical systems.
I am looking at this from the perspective of someone who has a few 10's of MB of data/writings on 25-30 year old 3.5 inch and 5.25 inch floppies. These are in formats for which I may not have an ability to extract the data/information without some significant effort.
The formats that have survived well for me over the last 30 years have been either open, or readable/writeable with open tools. The closed ones, not so much luck with.
[1]: http://mathics.org
[2]: http://mathics.net
UI is a major failure of open source - it can hardly ever achieve it, at least not well. Most of the popular open source programs have no UI at all.
Are the Python libraries precise enough for professional mathematicians? And do they deal with mathematical 'edge cases', a variety of inputs (formats, notations, etc.), etc.?
On one hand, I could say 'the author uses them therefore they must be sufficiant'. On the other, I've seen plenty of cases where the professionals were not careful about the tools they use (e.g, spreadsheets running critical, large-scale financial operations).
[1] https://reference.wolfram.com/language/guide/ControlSystems....
On the other hand, open source tools are catching up. Part of the SymPy project is SymEngine which is a reconstruction in C++. SymEngine is not as feature-filled as SymPy yet, but it flies. It's much faster than SymPy, and also Mathematica. SymEngine.jl works very well since by using the symbolic types in side of Julia you get both the speed (via function specialization) while getting a lot of free features through generic code. For example, Julia code is generic over number types, so if you call inv on a matrix of SymEngine symbols the built-in Base version will compile a fast version for SymEngine symbols and use it. So you have all of Julia's Base available along with Julia packages (yeah, you can put them in a neural net if you wanted to). So SymEngine.jl is the only thing that works when I need that speed, but I do have to keep Mathematica installed for its special handling of specific equations.
And I hate to say it but... the Mathematica notebook is really pretty. You put the exponents up, the fractions on top of each other...
yup. for quick and dirty type setting i preferred it to latex
SymPy is doing nicely, but it's decades behind Mathematica when it comes to symbolic computation.
By the way, if you want to invert a symbolic matrix, chances are that generic Gaussian elimination working with symbolic expressions isn't the best algorithm (it might even give wrong answers if zero testing is done incorrectly), and compiled code isn't going to change that. The state of the art in symbolic linear algebra uses specialized algorithms like evaluation-interpolation, modular computation, and all kinds of low level optimizations.
Mathematica is undergoing a compiler re-vamp, which may change that.
That's a video where they lay out how they want it to work, if you're interested and haven't seen it.
In that Jupyter files are just json files, I hope that there it will be easier to switch between the two in the future. Like [1], [2], and [3].
[1] https://github.com/gregsexton/ob-ipython
This is a weird jab at Julia. Open source software is woefully underfunded. Julia Computing was founded in the wake of Heartbleed where people learned that open source needs some kind of funding to keep developers alive (example article: https://arstechnica.com/information-technology/2014/04/tech-... ). Coming from academic backgrounds, the core contributors really had two options if they wanted to devote full time to Julia: either everyone gets an academic job while working on Julia instead of papers (lol), or band together to get R&D funding and use that to fund a life of open source development. They did the latter.
It's quite silly to even imply there's something nefarious that can go on here. Their main product is the language. They can't sabotage that without sabotaging themselves. They may have some priorities swayed, just like how any other individual who's working on open source is doing it for their own reasons. For example, IBM funded them to add PowerPC support, and what do you know Julia works on PowerPC. Is that so awful? With this funding model, what ends up happening is you have a large group of people who dedicate their lives to developing open sourced code for automatic GPU compilation, machine learning libraries, etc. along with compiler support for optimizing scientific computing. Because of this (and other reasons), Julia ends up having a much stronger governance which is one reason why its development ends up being more active. And this activity in turn makes its project more democratic than projects like CPython or Jupyter which have been larger projects for a longer time, but with less contributors (Julia's 686 vs CPython's 524 vs Jupyter's 330).
And most of the Julia contributors aren't even part of that company! Many are academics. A lot of the funding is through NumFOCUS, a non-profit which also helps projects like Jupyter, matplotlib, etc. which the author is for! (And they are great projects as well!)
So while I am happy that the author is pro open source, I think it's necessary to point out that this open source outsider view is both wrong and dangerous. Saying that you love the purity and despise anyone who gets to make a living from it is harmful! Open source is a labor of love, but it has also destroyed many careers. I think society has this view that open source (mathematical) projects are "funded" by academic careers, but even creators of popular projects like SageMath have publicly noted that open source is harmful to academic success (https://escapethetower.wordpress.com/2016/06/13/creator-of-s...).
Instead of being against funding open source contributors, I would like to see the author promote funding for open source. Paul Romer is a leading economist. He has the power to proclaim that open source matters for academic careers and push for it to be put on equal footing with papers for grant applications in his field. People like him should be advocating for jobs dedicated to open source development, not scoffing at the supposed impurity of someone being paid to develop a public good. Someone at the top of the academic hierarchy should start a change and make the development of public tools as valued as the development of (non-public) publications.
JupyterLab is in beta and is intended to replace the current Jupyter front-end.
I have seen table of content, but those have been generated by a big block of javascript.
I used this as a starting point and modified it for my own purposes: http://blog.juliusschulz.de/blog/ultimate-ipython-notebook
The initial solutions may be proprietary, and financed by investors. They have a business model so of course they don’t give everything away for free.
With time, enough people get together to build an open source alternative. And then like a snowball it eclipses everything proprietary that went before it.
What would a world look like that didn’t apply Capitalism to ideas?
One where companies couldn’t sue one another for Intellectual Property infringements. Like Waymo suing Uber.
One where self driving cars can incorporate improvements made by any other self driving car instead of putting people at risk reinventing the wheel.
Where the long tail of drug research leads to something.
Why would people release their findings? Because if they don’t, others will. And then they won’t get that small measure of input and control and attach their name to it. Jonas Salk is an exception in the biomedical field. Albert Einstein is the norm in physics.
The alternatives to Capitalism do not have to be Socialism. They can be SCIENCE. OPEN SOURCE. WIKI.
Collaboration instead of Competition.
I would like to see the same in web browser engines. WebKit instead of IE. And so on. When that happens, we all win.
Yes free software is good. Software, like knowledge, does not have to be scarce.
> The initial solutions may be proprietary, and financed by investors. They have a business model so of course they don’t give everything away for free.
> With time, enough people get together to build an open source alternative. And then like a snowball it eclipses everything proprietary that went before it.
I wish it did. The most obvious counterexamples that come to mind are Microsoft Office and Adobe Photoshop. With maybe the recent exception of Krita for the latter case, there are no known open-source alternatives that don't suck hard compared to the propertiary applications I mentioned.
I'm not really sure why that happens, but open source doesn't seem to scale well when building large end-user applications.
I also think that open-source needs some sort of end-user facing consulting, the “last mile” at it were. A Wordpress expert or hosted service setting up your blog. A browser like Chrome that uses WebKit or Blink underneath.
But the underlying free platform beats the proprietary ones in the end. Let people build their little competing businesses on top of it if they want.
In the world I’m talking about (collaboration via contributions) everyone would get UBI and be able to not contribute anything.
The straight one to one fight is between sagemath, or their rebranded cocal site, and mathematica. In terms of ability to do things sagemath is the glue between all the amazing open source math and science software that has been written over the last 50 years.
But in terms of presentation by far the best is org-mode with sage, julia and everything else tied in. One text file that can be emailed around, put under version control, and can speak two dozen computer languages with babel, and has latex support for both pdf and html output.
It's extremely interesting how they have managed to create an all plain text environment that is in some ways[0] better than the current notebooks. Of course it's not as user friendly and needs quite a bit of polish, and a major refactor to separate the above use case from the tangle use case.
The tangle part of org is even more interesting as it's probably the first fully literate system that allows you to completely document everything about a software project. And since it's language agnostic I do mean everything from install, compile, connect to the external databases, export the documentation in every format under the sun and whatever else you can think of (including docx through pandoc).
I'm going through the source code slowly to get it streamlined, but the code is suffering from the typical lisp syndrome of "Everything is implemented everywhere."
[0] Being able to run multiple languages on the same org file, controlling environments for scripting languages, compiling on page for non-scripting languages.
The basic strategy is to provide a new Python library ("the sage library" [1]), which is a layer between the user and the other dependencies. E.g., instead of just saying "use Maxima for symbolic limits", we have our own notion of symbolic functions (implemented using Ginac), and automatically convert them to and from Maxima when doing symbolic limits. The user never has to know anything about Maxima. When we first implemented this, it just used interprocess communication under the hood, but (thanks to Nils Bruin and others) we now use a direct C library interface builton top of ECL (embedded common lisp). This "new Python library", which isn't so new, since I started it in 2004, is now nearly a million lines of code.