Wolfram Mathematica

Michael E2

12:47 AM

@Young CloseKernels[k]?

Young

1:06 AM

@MichaelE2 That doesn't seem to remove them from the Kernel Configuration Options window

Michael E2

1:28 AM

@Young Sorry, my mistake. For some reason I mistook it for the parallel kernels window.

Karsten 7.

3:02 AM

@Young Are you talking about any setting in the following window?

All these can be change in principle using commands evaluated in a notebook.

Which settings do you need to change?

Karsten 7.

3:26 AM

Maybe Parallel`Palette`Private`autoLaunch = False already solves your problem.

The default is

Parallel`Palette`Private`autoLaunch = Automatic

J. M.

3:51 AM

@Nasser pretty sweet, innit? The Mellin transforms, too...

@PatrickStevens If you want to try stuff out before downloading, see this.

3

J. M.

5:29 AM

Somebody should probably implement this in Mathematica.

kirma

7:03 AM

Hrmmmm. If I understand it correctly, new neural networks functionality is highly numerical, you can't really extract symbolic presentations of computations from the presentation (without own code at least).

For study of rather simplistic toy networks this could be useful.

Nasser

7:49 AM

I got my first kernel crash in V 11. I'd like a price if I am the first to be able to do this :)

After about 6 hrs running.

Yves Klett

8:23 AM

@Nasser try not to think about the ID procedure too much. I need to fill out the very same form every time for my home-use license for every freaking release. This is quite horrible, as waste of time goes.

Martin John Hadley

For the experienced user... wolfram.com/language/new-in-11/for-existing-users

J. M.

@kirma, since you seem to be the one with most 3D-printing experience among us: once you get a copy of 11, can you talk about how useful the now built-in stuff for 3D printing is?

Martin John Hadley

There's a lot of stuff there that isn't new but I fell off this horse a little after 10.1, is wolframscript actually new?

J. M.

8:46 AM

@MartinJohnHadley Yes, some of the announced functions were actually introduced in 10.x

BoLe

If somebody is using the new version 11, can he or she checks Length[Names["System*"]]`? I'm interested, I have 10.3 and there are 5646 symbols.

J. M.

@BoLe, too bad you deleted it, but: WolframLanguageData[] // Length gives 4837, and Names["System`*"] // Length gives 5900.

Szabolcs

9:03 AM

@MartinJohnHadley In a way it's new. It's different than what we had in 10.x. Different command line options, new features such as cloud evaluation, etc.

Shutao Tang

Does Mathematica 11 using the new programming font?

kirma

@J.M. I installed v11 yesterday, but haven't done anything seriously with it yet.

My wild (and not so informed) guess would be that as long as you have relatively sane polygonal/mesh models, results are good. Somehow I fear that constructive solids geometry (even between semialgebraic regions) still lacks code to discretize sharp edges neatly enough not to be a quality hazard.

At least I didn't run into anything that would have advertised big improvements on that front.

One thing I didn't see being mentioned is working with various material design rules. I'm not surprised of that, though, because those are process technology specific and not necessarily at all easy to reason automatically.

Szabolcs

@kirma It has mesh repair functionality.

kirma

That is, things such as minimum thickness of "wires", minimum sizes of holes, requirements of no closed internal cavities and such.

I believe mesh repair functionality is a good thing assuming your models are mesh or polygon based. :)

I'm concerned of problems like this:

13

Q: Controlling quality of discretized region meshes

EDIT: Impact of this problem has been considerably lessened in v10.2. Although new version is still not performing analytical discretization at sharp corners (there are jaggies), evenness of sampling seems better and the MaxCellMeasure option is actually obeyed in the expected manner. This makes...

The jagged edge which should be relatively easily (although with untrivial computational cost, as the model geometry gets more complicated) avoidable by intelligent discretization.

Szabolcs

9:19 AM

@kirma It doesn't repair the meshes I care about :-(

kirma

:(

Mariano Suárez-Alvarez

All I see is that finally there's a Smith normal form thing built in, after all these years!
tears_of_joy.gif

kirma

Yep, quick look at mesh discretization would indicate that there are no huge improvements on those jagged edges.

Martin John Hadley

9:41 AM

@Szabolcs thanks I thought it was new, if it's easier to talk to the cloud with wolframscript I might have a look into it. I don't have it yet, are there still the issues with multigraph? You added a bug tag for me on here mathematica.stackexchange.com/questions/92014/… that I was going to update when I get my hands on m11

Szabolcs

9:59 AM

@MartinJohnHadley I think that's essentially broken design and I don't know how they will work around it ... It's not fixed in 11. I have the impression that the graph stuff is stagnating. There are few fixes that I noticed during 10.x and 11.0 and many known wrong-result bugs are still unfixed.

Szabolcs

10:22 AM

Has anyone played seriously with the neural network functionality already?

J. M.

10:33 AM

@MarianoSuárez-Alvarez SmithDecomposition[] has been available since 10.2; unfortunately, like HermiteDecomposition[], it is only restricted to integer matrices. You still need to write for the polynomial case if you need it.

Szabolcs

10:55 AM

@Searke Could ImageIdentiy be done using the v11 NN tools in practice?

kirma

NN becomes interesting the moment it can a) use existing weight files with ease, say VGG-19, and b) training can be delegated to cloud (AWS GPU instances, Google Cloud Machine Learning) without licensing or technical hurdles.

Recurrent neural networks would be also welcome, but supposedly they're coming eventually...

I'm not planning to buy large-memory GPU for experiments on NN, but I gladly buy GPU hours on AWS spot pricing.

WRI might choose to resell that capacity branded and streamlined as their service, but if pricing is tenfold or more in comparison to average spot price of that utility, good luck trying to find paying customers. I could pay for it even as a home user if it isn't obvious I'm getting overcharged for something I could do myself for a fraction of cost.

Szabolcs

11:19 AM

@kirma Yes, that's why I was asking about importing caffe data yesterday

kirma

I guess "compiling", or rather, script generation for Caffe/TensorFlow/whatever they're called shouldn't be particularly hard from NN definitions, but it's an interesting question if WRI is eager to provide that functionality. I bet it would improve attractiveness of NN on Mma quite a bit.

user21

1:30 PM

@Szabolcs you should send these meshes to the support - not my construction site but if these are available then there is at least chance that the repair can get improved.

@kirma please send a request to the support the more people request this the higher the chance that things will more up in priority.

Searke

1:46 PM

@Szabolcs If I had to guess, I'd guess acquiring the data to train it might be the limiting factor. I don't really know.

Edmund

New Messages Interface in version 11. I think I am in love.

Searke

Training in the cloud is something they're looking into. No idea about pricing for that.

Competing with AWS pricing is (in humble opinion) kinda like asking a store with hand crafted items and good customer service to compete with Walmart's prices.

kirma

@user21 OK! :)

@Searke "Competing" is one thing, being reasonably more expensive with good integration is another.

I'd assume components of the cloud infrastructure would be running on one of these providers anyway, and it's not that hard to make things like NN training relatively tolerant to spot pricing aborts.

Szabolcs

@user21 I have already sent them during the beta period. Of course not every mesh repair technique can deal with everything. I tried multiple tools, some could repair it and some couldn't.

@user21 I was just hoping that maybe v11 final was improved in this respect. I noticed a few fixes between the last beta and v11 final, so I was hopeful.

Szabolcs

2:07 PM

I'm going to drop this here because it is already at the bottom of the main page ...

2

Q: Set spell checking language

Version 11 brings real-time multilingual spell checking. The linked page says, The language is automatically detected as you type, but it can also be set manually. How do I set the spell checking language without changing the user-interface language?

front-end notebooks spellchecker

Michael Hale

2:24 PM

Downloading v11. Due to some craziness I now have a license for Mathematica, Mathematica Home Edition, and Mathematica Home Annual Edition on my products page. The upgrade was available for all but Mathematica Home Annual Edition, which confirms @kirma 's suspicions from a while ago.

@Edmund I'm also glad they added stack trace to messages. I actually had some very pleasant experiences with the Wolfram debugger recently once I realized that I needed to evaluate all of my function definitions in debug mode. Then everything worked as expected. Previously I would switch to debug mode and then only evaluate the function call I was trying to debug. Then I would get frustrated that break points wouldn't work.

The documentation should mention that you need to evaluate all code where you want to place a break point in debug mode.

Michael E2

2:46 PM

@user21 I have a question, if you have a moment, about the NDSolve plug-in method foo[data_]["GetDenseOutput"[args___]]. Or I could ask over email, if that would be better.

user21

@MichaelE2, either way is fine.

Michael E2

@user21 I'm trying to write a method that returns dense output in ChebyshevT series. I stumbled on "GetDenseOutput", but I can't figure out how the return value should be structured. Can you tell me? (Is that enough info.)

user21

@MichaelE2, I don't know off hand but I'll have a look if I can find something.

Michael E2

I always get a NDSolveValue::baddo message, if that's any help.

@user21 If there's an example, I could probably work from that.

user21

@MichaelE2, I have something but that's a bit long to post in chat here, if you quickly ping me at ruebenko AT wolfram.com then I can send the little info I have.

Michael E2

2:53 PM

@user21 Thanks.

Michael E2

3:34 PM

@user21 Thanks again. It looks helpful. So far it's not working, but let me play with it. Maybe I've another error somewhere.

user21

@MichaelE2, maybe the information is outdated. If you can not figure it out ping me again and I'll try to find someone that knows.

Michael E2

@user21 OK, thanks.

J. M.

4:22 PM

@user21 hey, can you cc me on that?

I'm looking into Chebyshev stuff too, so this is of great interest.

Thanks!

Yup, multistep and Bulirsch-Stoer would of course be the most obvious users of the framework.

Nasser

6:01 PM

I think one of the killer apps for Mathematica is the show step-by-step (S-b-S) solution for problems, like integrations, solving differential equations and many others. Wolfram alpha pro has this feature (even though it is limited, but still useful). So I was wondering, why WRI would not have this feature in Mathematica itself also? It will increase sales. I know many students who only use Alpha just for this feature.

chuy

@Nasser It is in Mathematica, via the WolframAlpha function

J. M.

6:33 PM

...I slightly dread the day that a Verbose option will be added to Integrate[], DSolve[], D[], etc. :|

Nasser

6:44 PM

@chuy I see. Will try it then. But this is strange, since step-by-step on Wolfram alpha is only for those who have PRO version. I bought PRO version just for this feature. I had no idea it is free also for those who have Mathematica!

@J.M. I think an option should be add in Mathematica, such as ShowSteps->True to many functions. This is what Rubi integration package does. If one sets the ShowSteps flag on, it will show the rules used. Something similar should be build-in Mathematica itself. THis is a very useful features and will make CAS system much more useful to learn from

For me, the show step by step feature, is much more useful than all this cloud and connectivity stuff that WRI is spending time and resources on.

J. M.

"...Mellin transform of integrand, express as Meijer G function, use Slater's theorem, convert back to hypergeometric function..." Hmm...

I guess for very elementary stuff, sure.

Szabolcs

7:07 PM

@Nasser It works, just start the line with two = signs.

Michael E2

7:19 PM

All the built-in methods that I've tracked down so far have a method defined like this: ``NDSolve`Extrapolation[MethodData[__, Pattern[ExtrapolationDump`data,
Blank[]]]]["DenseOutputQ"] := False;`` AFAICT, `"GetDenseOutput"` is not called in this case. The dense output data is created internally somewhere that seems hidden.

Sham Says

7:33 PM

Anyone know if Mathematica 11 comes with a English grammar checker function of sorts?

Searke

@Nasser Not exactly. That is not what Rubi does. Rubi is simply opening itself up and showing you what steps it took to solve the problem. The step by step feature is trying to construct a way to solve the integral that a highschool student would understand. It is not showing how the actual Integration was done. It exists in Wolfram|Alpha because that's where students use this kind of feature.

@ShamSays TextStructure?

Otherwise the answer is no.

The notebook interface itself doesn't do grammar checking. It does spell checking.

J. M.

As Searke says, what Mathematica actually does is very much different from how (normal) people do it. What I quoted below you, @Nasser, was only a slight exaggeration of what might happen behind the scenes.

In effect, having that option would be tantamount to making Mathematica solve a problem twice: the computationally convenient way, and the "paper" way.

Szabolcs

7:56 PM

@Searke Yes, but is there any reason why this functionality should depend on the cloud and not be part of Mathematica itself? I thought this was what Nasser was asking.

BTW Mathematica 11 is 7 GB on Mac which is starting to be a lot ... I use HFS+ compression on it to manage to fit several versions on my laptop. It saves some 40%. That's probably because much of the space is taken up by documentation, which compresses very well.

I assumed that some stuff is left in the cloud because it would simply be too big for Mathematica.

Searke

It's not free to move functionality into Mathematica

it has to be balanced against the cost to do it and maintain it and the amount that it will get used.

Although, I will admit this gets asked fairly often and I don't think people have adequately communicated to the relevant people the amount of interest in it

Karsten 7.

@Searke Could you please ask the person responsible for the MathematicaClosingDialog.nb to replace Scaled[0.33] with Scaled[1/3]. I already did it for my installation, but wouldn't like to do it again for the next release. It causes some funny behavior during the first MouseOver.

Karsten 7.

8:20 PM

Btw, it looks pretty cool and fits very well into the new Windows 10 Dark App-Modus using a black background.

Kuba

9:18 PM

why

FileNames["D*", "c:\\"]

generates

General::cdir: Cannot set current directory to PerfLogs. General::cdir: Cannot set current directory to Recovery. General::cdir: Cannot set current directory to System Volume Information. General::stop: Further output of General::cdir will be suppressed during this calculation.

?

chuy

9:35 PM

@Kuba I wonder if its a pernissions thing

xslittlegrass

9:54 PM

@Szabolcs I also have several versions, and they take a lot of disk space. How do you setup HFS+ compression?

Transcript for