Has anyone noticed an issues with EventHandler(s) in 11.3 on Win7? I have an application which users report is unusable with 11.3 but works fine in 11.2. No problems on OS X

Does anyone undergo sluggish Mathematica frontend in 11.3 after doing a bunch of plotting and calculations?

@b3m2a1 often bugs arise for a reason - interesting corner cases, pedagogical progressions, dependencies, language choices and design, code coverage, coding ticks etc. Hence I imagine classifying these would be useful for automating the creation of unit tests, documentation examples, coding/design development etc - all things naturally combinable in a package-management system.

11.3 has the problem as Mike and others are claiming. Many things works better in 11.2. For example importing JSON is broken Often it fails with json that can be easily imported with 11.2.

I hope 12.0 is not more unstable and sluggish than 11.3.

Hello, is it a known problem that when exporting a ListPlot as PDF with low ImageSize the coordinates of the points become really inaccurate?

E.g. ListPlot[E^# & /@ Range[0, 2, .02], ImageSize -> 200], the horizontal spacing is very inconsistent. And the y-coordinates are off too (more noticeable with other functions).

Is it appropriate to ask a simple meta etiquette question here, or should I create a formal meta question for a simple issue?

@BenKickert First rule of the fight club: Don't ask if you can ask. Simply shoot and if it should go to Meta, someone will let you know :)

@pcworld The image size should not influence quality when you export it as PDF. This is how it looks on my Mac with 11.3

@halirutan: I just noticed that it happens only when exporting via GUI (right click -> save as), but not with Export. Using Linux with 11.2. Did you use Export or GUI?

@pcworld I used Export.

This is the GUI variant btw:

@pcworld Christ.. yep, looks here equally awful when I use the GUI. This should definitely go to [email protected]

In my question about finding distance along a GPS route, the answer given is very elegant and gets me 90% to where I need to be. However, it doesn't account for instances where measuring distance would double back on itself and introduce error. Using the code from the answer as a base, I am able to come up with a code that gets me exactly what I am looking for.

@BenKickert The answer is yes.

Is it better to create my own answer and credit the original answerer as the inspiration, or to edit the original answer and add section that says "To avoid error with doubling back you can...". Or somethign else? I want to give credit where it is due, but also ensure the accepted answer fully addresses the question. Thoughts?

@halirutan I'm not subscribed to their support so not sure whether I can report it. I was surprised I didn't see a question here about this issue.

You go ahead, post your own answer, thanking the other poster one as much as you can :)

And then accept my own answer?

I do love the structure of StackExchange because it keeps the questions and answers close together, but there is certainly a learning curve with getting things to that point. Thanks for the input.

@BenKickert Well, you can do that but people will see your answer as well if they are interested and it might be nice to rather accept the one that tipped you off 90% of the way.

@pcworld You don't need to subscribe to something. You simply go to Help -> Give Feedback.

How are duplicates of questions on SO (not MMA.SE) normally handled? I'm asking because of this question

@pcworld If you get answer from them that it is indeed a bug and not a feature to make plots more interesting, would you consider asking a question on main about it, tag it with [bugs] and insert a bug-header like it is done in this question?

@LukasLang You could vote to close it and use "other reason" where you explain that this has been answered on SO and you provide the link. OR you search carefully if we have a duplicate already on our site.

But I'm not entirely sure. @Mr.Wizard Can you comment on the question of Lukas?

@halirutan Will try to remember if they respond. Should the workaround go into the question itself or into an answer?

@pcworld If there is a workaround, feel free to answer you own question.

Ok. Well the workaround would be using Export[]

@pcworld Yeah.. :) I was thinking about a GUI workaround.

Curios about people's opinion here. This question brought up the fact that generation of Hadamard matrices is shockingly slow in MMA compared to e.g. Matlab for essentially the same result. Indeed, @Coolwater proposed a one-liner that beats the built-in by a few orders of magnitude.

I understand that the built-in returns a slightly different version of the matrix (see comments too), and that it the MMA result is calculated at arbitrary precision by default, but isn't the most common use of those matrices likely to imply machine-precision anyway? Am I missing an obvious design choice here whereby the current approach in MMA is preferable?

Perhaps the docs could at least mention that by using a different method and working at MachinePrecision you can get a result 100x faster.

@MarcoB If you look at the Twitch videos, something that stands out for me is that the naming of functions is discussed a lot, whether something would be intelligible for a high school kid is discussed a lot, but there is almost no mention of performance, and virtually no analysis of whether an API would hold up during practical use (or in anything longer than a one-liner typed right during the meeting).

Design is clearly not driven by real use cases.

@Szabolcs Yes, it's been frustrating lately even in my own limited practice. And this opinion has been supported by examples even in the last few days. Only recently Henrick Schumaker wrote a compiled version of quicksort for Reals that outperformed Sort! I found that shocking as well.

In fact, I had initially (somewhat snarkily) commented that you could hardly beat the built-in on something that basic that must have been thoroughly optimized. And I was proven wrong...

I thought I could rely on Wolfram to optimize their implementation of basic well-known algorithms such as sorting as much as possible, so I don't have to worry about those when I write high-level functions. I understand that Sort does more than sorting reals, but still!

If it makes you feel better, I as a noob cannot even figure out which part of my code is slow, or whether it's the builtin functions that are slow.

Mathematica is cool as a magic black box, until the magic doesn't work anymore or is very slow.

@pcworld :-) But see, I rely on MMA as a magic box as much as you do. Indeed, that's the crux of the problem. I want to be able to assume that, when I use a built in, I automatically get the best performance that can be had from the system. I don't want the nagging feeling that I could perhaps do (MUCH!) better by rolling my own.

In truth, compiled functions "á la Schumaker" are not the answer for me either. I use Mathematica precisely because I don't want to write loopy C code :-)

@MarcoB But this is exactly where the slowness comes from. You can Sort all kinds of things and Mathematica needs to make sure such basic functionality always works. Therefore, it has to take care of many corner cases until you finally reach the implementation of sorting a list of real numbers.

The other day I learned that NMinimize calls the function multiple times with the same value. Great for performance too...

@halirutan Undoubtedly. I would advocate, then, for the underlying specialized functions to be made available at top level in a documented way. Perhaps an even more striking case is Import, which has to deal with any old garbage I throw at it. When one can find the specialized function that does the importing one needs, though, that's typically a lot faster. I guess it would be nice, sometimes, to be able to bypass the checks and balances when performance is paramount.

There's lots of similar questions, but none have had a working solution for my use case (at least I couldn't get it working) and I'm unsure whether this would be duplicate.

@pcworld No, I don't see a direct solution. I could go into detail, but the core of the problem is that a plot with legends is basically a grid containing the graphics and the legend separately. So when you set ImageSize explicitly, you plot will be larger by the amount of the legend. The only reasonably easy solution might be to export the graphics and the legend in separate PDFs.

If you only export the plot, you can calculate the image width to the millimeter and that's what I do for my LaTeX documents. If there is a legend, I simply take care that the plot has the same size as all my other plots and that there is enough room for the legend.

For some reason i no longer get acknowledgement nor responses when I use the form that pops up when one does "Help > Give feedback". Have any of you received or not response lately from Wolfram Support in that way?

@rhermans I wrote a few days ago and got the usual quick reply.

@halirutan mm... it's me then. Thanks.

@halirutan Thanks for the hints. I'll probably resort to that or inline legends.

I don't suppose anybody knows what the "AudioFeatures" feature extractor is built on? It's very effective, but I would love to know what it's actually doing. It's mentioned on the FeatureExtract and related pages, but has no doc page of its own.

Unfortunately MachineLearning$FeatureExtractorInformation` has little useful information...

I suppose it points to the last layer of some NN being a likely solution (which is what I assumed it would be)