@ChrisK, thanks. Let me know what you think once you had a chance to look at it. Feedback is always welcome.

@user21, can it solve equations with a PDE coupled to an ODE?

@JanPöschko I was hoping you'd see that, thanks a lot for the information!

Shocked that the app icon doesn't have a new spikey. Shocked, I tell you

@KraZug, no :-(

Can confirm the 12.0 FE also likes to hang and crash -_-

But then again it's impressive it's not worse than 11.3 given that it's wholly rewritten

It feels a lot more responsive to me. I look forward to testing the limits of what it will display before it crashes

Yeah some parts feel faster (like scrolling) but others are a bit slower, especially when doing heavy duty Dynamic manipulation in the FE.

It also crashes on more display corner cases, I found

Like calling NotebookOpen on a .xyz file made it hang, and trying to display a ~400KB blob with a weird head from an error made it crash

I'm gonna report these when I have some free time but I have not yet been able to sit down and reproduce the crashes

Updated the ad. Could I please get a couple of upvotes to bump it over the threshold?

@Kuba Sorry for the delay - here is one example (you can find another one, together with a few of my attempts here: Notebook):

cell = Cell[BoxData[
   RowBox[{
     RowBox[{"f", "[", RowBox[{"a", ",", "b"}], "]"}],
     "+",
     RowBox[{"g", "[", RowBox[{"c", ",", "d"}], "]"}]
     }]
   ], "Input"];

Desired output would be:

"f[({
 {a,b},
 {c,d}
}),3]^2"

? there are no line breaks in the cell, nor () etc.

So I am not sure where does the output come from

@LukasLang ^

@SjoerdC.deVries I think my download was something like 35 MB/s, which would be closer to 300 Mbps... it is not unimaginable that it was actually bottlenecked by my home wifi (wifi router behind a wall). Usually I have been using phone mobile hotspot for this download (it's great to live in a country where data contracts tend to be unlimited), so it's hard to compare if it has improved since <100 Mbps long term speeds on 4G network definitely make it more painful.

(Operators definitely deprioritize you if you download gigabytes of data over mobile network at full throttle.)

@Kuba Good catch - I was too preoccupied by the fact that TextClipboardType->"Package" seemed to give something acceptable... "f[({{a,b},{c,d}}),3]^2" as output would be nice

@LukasLang I think you pasted cell1 instead of cell2 so the result does not match

The goal is to be able to convert the resulting string back into a box structure using FrontEnd`UndocumentedTestFEParserPacket - so I'm looking for something like "Box structure" -> "Readable string" -> "very similar box structure"

@Kuba I think I'll go hide under a rock for a while - of course, you're completely right. For the cell structure above, the output would be "f[a,b]+g[c,d]". The output pasted above ("f[({{a,b},{c,d}}),3]^2") belongs to

cell2 = Cell[BoxData[
   SuperscriptBox[
    RowBox[{"f", "[", RowBox[{
        RowBox[{"(", GridBox[{{"a", "b"}, {"c", "d"}}], ")"}],
        ",", "3"}], "]"}],
    "2"
    ]
   ], "Input"];

@LukasLang as usual the devil is in the details. And that matters even more if you want to do a round trip. So here is another one problem: "f[({{a,b},{c,d}}),3]^2" the information about Grid is lost here, isn't it?

Do you want the string to be in InputForm or OutputForm, if the former, how do you want to preserve information about 2D typesetting of the original structure?

I assume InputForm because even though OutputForm can contain more information it will look ugly and the round trip won't be so much easier anyway

@user21, ok, thanks. Can it do PDEs with spatial derivative higher than 2?

@KraZug, no the FEM can not :-(

@user21, I guess that means that xzczd's method of lines functions hvaen't been made obsolete then

@KraZug, the method of lines will never be obsolete since that is the main mechanism to time integrate PDEs with different spatial discretizations like the FEM or the FDM.

yes, but it would be great if NDSolve did it automatically for more cases

New "StructureGraph" import element for NetCDF is really cool

@KraZug, I am not sure what you mean. Could you explain a bit?

GeoWithinQ still quite slow, unfortunately. One of my most-used but slowest functions

A reminder to everyone updating bug posts to use the standard header!

@Szabolcs Do we keep the bug tag if it's been fixed?

@CarlLange Yes.

@CarlLange our 'bug tracking' is for the community, not for WRI. And community uses variety of MMA versions.

@Kuba Makes sense.

"RawJSON" import broken for anybody else?

Hm, works again after a restart

@CarlLange Just curious: is it slow also for repeated application of the operator form? In theory it could build a reusable search data structure...

Also I wonder if, when using entities, getting one point of the loc entity and calling GeoWithinQ (or the operator) would be significantly faster and sufficiently precise...

Baseline polygon intersection test on a geoid sounds pretty nasty... and overkill.

aha, in the docs there's an example using Pick that might be a lot faster actually

Yes, OK, I am using it wrong

much faster.

(I'll probably post a question about this soon, since it is starting to be scattered over many messages already...)

Still pretty slow though, really. If I have satellite data of europe with some relatively high resolution (100k points, let's say, but even that is kinda low), getting only the points within dublin city takes a long time.

@CarlLange Interesting... not necessarily intuitive.

@kirma I have some alternative that converts to Region and does RegionWithinQ which was significantly faster.

oh, wow, it actually uses the network

I just tried above with 10,000 points and got GeoWithinQ::servermessage: Server message: Internal server error.

oh dear

ehmm....

I suppose that explains why it's so much faster to use the Pick version

It's sending all the points at once to the server rather than one-at-a-time

s/RegionWithinQ/RegionMember above

Sounds... odd.

oh man, I bet you RandomGeoPosition[region] also uses GeoWithinQ, so generating a large list of points within europe is also super slow

Also, haha, is the "Give Feedback" button in the Help menu on OSX accidentally bound to the "Why the Beep?" function?

Actually that menu seems mildly broken anyway

I can sign out twice, which is cool

@CarlLange Nice. I do understand that planar vs. geodetic membership tests are different, but that's quite a lot. :)

@CarlLange I see that too. :)

@kirma Yes, I think the big thing here is projections and so on. There's definitely potential for a lot of complexity in a GeoWithin function, but I kind of want PerformanceGoal or something.

Help menu seems really messed up. But there's no problem if you can't give feedback easily!

Reminds me of this:

assets.amuniversal.com/4bd6edc06cb801301d46001dd8b71c47

:D

I just sent in feedback via the online form.

any other Mac users tried out FindFaces or FacalFeatures? I consistently get an empty list returned - even if I evaluate the cells in the documentation. Windows version seems fine.

seems like an issue with the Paclets.

@kirkus Works for me, on Mac.

@kirma hmm. what macOS version are you on?

10.14.5 Beta 1 ... :I

Those examples do download a lot of stuff though.

yeah nothing tries to download most of the times I run it.

I've even whacked my whole ~/Library/Mathematica folder

Oh here's a interesting "new in 12.0" bug: my tab switching mechanism now breaks if there's no content in the notebook.

I think a NotebookDelete on the empty Cells[] is doing it...

doesn't seem to be a connection problem, AirTemperatureData, FinancialData, and AstronomicalData work fine.

support time I guess!

@kirkus Works for me just fine on Win 10. I get Rectangle[{85.5, 89.5}, {168.5, 185.5}] :)

aha!

clearing ~/Library/Caches/Wolfram took care of it.

Hi all, generating a large number of equations

I can generate them in Hold pretty quickly, but ReleaseHold[] during later use actually jacks up computation times

What gives? Or does it depend on what I'm doing? I want to use symbolic variants inside FindRoot so Compile doesn't seem the way to go

Selected tab on my Mac Mathematica preferences window is black text on (almost?) black background. Somewhat impractical.

Watching the "livestream" ... "... aand that's the world of holy polygons." Sounds religious.

Especially when the demo was presented with a five-pointed star.

Exploring M12.0's ability to interface with Python libraries: Using networkx from Mathematica

@Szabolcs nice writeup!

More like a quick and dirty cut and pasted from notes I made during the beta testing period. I hope it will be useful though.

In M12.0 the Python -> Mathematica conversion is done in a structured and extensible way.

Hmmh. I wonder how "PreferredGPU" setting under graphics RenderingOptions actually works...

What is still missing is the same for Mathematica -> Python. I hope to see it in 12.1! The groundwork is clearly done in the Wolfram Client for Python, but it is not currently being used by ExternalFunction.

@user21 Will do -- I'm always happy to see some attention to developing the Math in Mathematica :)

Sometimes I'm annoyed to high hell by the fact Mma likes to turn discrete GPU on for no apparent reason and not to release it. Especially on long intercontinental flights...

I think I'll give MATLink a spin on 12 and see how it goes

@kirkus I expect it to work fine, though I only tested on Mac.

@kirma I use gfxCardStatus to prevent that.

@Szabolcs I think it stopped working, or something. Or did it?

@kirma I think I am using this modified version github.com/steveschow/gfxCardStatus/releases

@Szabolcs Hmmmm.

Hi, is this not the chatroom to ask general mathematica questions that may not warrant a question on the site?

Should I be asking on the Wolfram community forums instead?

I've asked some questions on TeX SE and these were all very well received. Coming to Mathematica, Stack Overflow and here is in tremendous contrast with that. I don't think anyone ever even bothered with my question...

I just don't really understand

I've checked the meta but nothing significantly different to tex..

@1010011010 You can talk about anything in this chat, the chat description even says so. ("Other than the general guidelines in place, anything goes here.")

Be assured that you haven't done anything wrong.

If you want to increase your chance of getting an answer, the same principles work here as on the main site. Explain your question clearly. Share code. Provide an MWE.

@1010011010 Which of your questions were ignored here?

@1010011010 I think they're just a larger community than us

@Szabolcs First questions about procedural programming (other acc), then questions about symbolic integration, then questions about back-end and Integrate's option TraceInternal, and now questions about list structures. I'm just trying to understand if it's me or something. This problem has been so stubborn I can feel my cortisol levels rising in real-time

Can you link to something specific?

I'll have to take an hour to find it again, these questions I have asked over a 4 month stretch ..

There are no questions linked to your profile.

Sorry to be blunt, but I hope you did explain the issues clearly, stated a question explicitly, linked to relevant materials and addressed comments/questions below the post. In this discussion you did not point to a concrete example even after I asked about it twice. If something similar happened on the main site, that would explain why your post was ignored.

@1010011010 If you don't receive answers here, why not ask on the main site? very many users don't use this chat room. My apologies that I didn't respond to your query earlier, but I know absolutely nothing about your area of concern. I would strongly recommend asking on the main site if your question doesn't get any traction here.

Or did you mean questions asked within this room, not on the main site?

I believe they meant in this chatroom. chat.stackexchange.com/transcript/message/49959456#49959456

In my opinion, this chatroom is good for answering questions if your question is of a complexity level like "I've forgotten the name of the function that does X and I can't find it, does anyone know it?" I think in general any detailed questions at all are better suited for the main site.

Far more users see questions on the main site (for instance, kglr, who answers 50% of the questions on the site, has never set foot in this room) - and the point system doesn't apply within this room. And the history of the room is very hard to reference in the distant future or for other users, as opposed to a question on the main site (even if it is closed for being a simple mistake, for example)

@Szabolcs: How do you find the performance of the new 12.0 MMA/Python interface? I have seen your criticism of the ExternalEvaluate (?) interface on Wolfram Community saying it was too slow for general work. I am looking at your description here on SE of ExternalEvaluate for networkx and it sounds like you are more optimistic. Would the 12.0 ExternalEvaluate for Python be able to pump through thousands of calls to Python and back in a reasonable amount of time? Or is there a lot of overhead?

Is wolfram script supported by wolfram language?

I was looking for something like:

<< script.wls arg

but I think I need to RunProcess?

@Kuba not sure about passing args, but: reference.wolfram.com/language/tutorial/…

@kirkus I am aware of that tutorial but I failed to find an answer there. Or do you want to point me to something specific there?

@Kuba no, sorry, just pointing out that generally it seems it should be supported, except I didn't see anything about being able to pass args using the Get/<< syntax

no need to be sorry :) thanks for initiative

somehow whenever I want to use a feature I didn't use before, it does not support what I want from it :)

nvm, have to go :)

@berniethejet I have not played enough to give a good assessment, but I think it's going in the right direction and I can't wait for when the Mma -> Python conversion will also work in a structured way.

I don't know about thousands of calls because of latency, but transferring large data from Py -> Mma should work much better in 12.0 now

Thanks Szabolcs. I'll take a look at it also, but still haven't got around to downloading 12...

@berniethejet Things like numpy arrays are about 30x faster in v12

And there is added support for data types like PIL images:

I wish I could give examples of anomalies to FindAnomalies

@CarlLange doesn't AnomalyDetection kind of let you do that?

@kirkus I guess. It lets me give it only good examples, which is something, I suppose...

It's also extremely slow with images of any dimensions > like 100x100. I suppose if I had a GPU it would be fine 😢

@CarlLange yes, my next MacBook may have to have a GPU at this rate.

How are you liking the new frontend for Mac?

@kirkus Sadly, you'll have to get one from 2013, which is the last time apple shipped an nvidia GPU.

or an external one!

I'm really liking it! It feels a lot faster, which I'm really appreciating

I've yet to crash it, which is a streak I'd like to keep going :D

same here, been really solid, even thru the prereleases

must have been a lot of work to rewrite

I wonder if much Metal is in there, or if they just went with all OpenGL for now

right now I'm trying to find practical uses for NumericArray.

Thanks @A

Thanks @ArnoudBuzing. I guess there are a few components of the roundtrip overhead: conversion from MMA to Python types, transmission, reception, and deconversion, is that right? So I guess the conversion/deconversion of the arrays is the speedup you are talking about, is that right? That is very good news.

Is NumericArray what PackedArray used to be?

no, they appear to still be different things.

yeah I've been trying some naive examples, basically using it instead of PackedArray, but it's not really there yet.

almost seems more designed for getting data in and out of NetEncoder and NetDecoder

@berniethejet this is a good 1 hour overview of all the improvements: youtube.com/watch?v=YXGyrkKQuVo (it's from the 2018 developer conference in Champaign, Illinois)

@CarlLange NumericArray is more like the documented successor for the undocumented RawArray(but the syntax is different and they are not quite the same). Functions which used to returnRawArray (for example RandomImage[1,{10,10}]//InputForm) now return NumericArray.

@ArnoudBuzing Aha, I see! That's good to know. It's a bit strange that PackedArray isn't documented yet considering how much of a speedup it can give you (particularly with regard to Parallel and friends)

s/strange/sad :)

@CarlLange It is documented to some extent, reference.wolfram.com/language/Developer/ref/PackedArrayQ.html Tutorials dealing with it will invariably mention it. But yes, there could be a better description in the docs. Packed arrays are very different from NumericArray because they are transparent. They are observable merely through performance measurements and a few developer functions. NumericArray is an expression type in its own right (mainly useful in conjunction with LibraryLink)

Correction: "Tutorials dealing with it" -> "dealing with performance"

@ArnoudBuzing But only way way (Py->M). Looking forward to the other direction.

For me, ExternalFunction/ExternalValue are one of the big highlights. They increase the utility of this system tremendously.

@Szabolcs what do you mean with 'the other direction' here?

@ArnoudBuzing Take a WL packed array, create a Py numpy array from it.

@ArnoudBuzing For example, send it as an ExternalFunction argument.

Bit of fun, my sister, girlfriend and I trying to sing to a scale:

PitchRecognize is cool

(And so is SoundNote)

also, love how responsive CloudDeployed plots are. So performant - even more so than the frontend