@C.E. This is 3ds export and rendered with Blender.

@yode what hardware are you using to test your style transfer net? i am buying a laptop in the next day or two to get a proper GPU, not sure what to shoot for. my goal is to use that exact same code (keeps totally locking up my mac!)

@ThomasLackner That is in itself a contradiction :) Bying a laptop to get a decent GPU doesn't really work.

@halirutan 1080 w/ 8gb should be enough to run demos, dont you think? not super serious stuff and im ok with slow results

What is your budget?

@halirutan $2kish i guess. there doesnt seem to be much above that anyway, except in huge form factors.

@ThomasLackner You mean they are pressing an NVIDIA Gtx 1080 into a laptop?

@halirutan yes.. search amazon. various models at all price ranges. i wont post those links here. MSI, Gigabyte, ASUS

@ThomasLackner I only see the "M" variants for laptops. Let me check amazon..

@halirutan amazon.com/MSI-GT73VR-Titan-Pro-201-I7-6820HK/dp/B01IS33J76/…

@halirutan according to benchmark sites, 1060 and 1070 are comparable performance. i dont know everything about this, hence my question in the first place. trying to sort it out. need to work faster

@ThomasLackner If it is indeed a real 1080, then go for it. You would do fine with a 980 as well.

What you have to look for is (a) the available cores and (b) the amount of GRAM in general.

@halirutan hm, ok then I see the problem.

@C.E. I have two things in mind:

1. A decent 2D representation of a 3D plot that contains vector graphics and not a rastered version of the plot.

2. An export of a plot into a format that can be rendered in high quality as you see in my example. For this, we would need to recreate the box, axes and ticks, as they are never exported. As you can see in my example, the color material of the surface is already exported.

Sounds good, I would particularly like (2). Would be useful for posting graphics on the web.

Since a few years, I'm really interested in number 1, but this seems an impossible goal. The problem is, that you can project and export all polygons, but all mathematical software I found always exports "all" polygons. Even the ones that are not visible. This leads (and has lead in the case of Mathematica) to incredibly large PDFs when you additionally color the surface.

It is (I believe) the main reason, why Mathematica nowadays only exports rasterized 3D plots.

The problem behind is that each Polygon of the surface has its own VertexColors according to the ColorFunction.

ok

This here is an example of matplotlib

Nevertheless, the PDF is small because they use not so many polygons and you have only one color per face, not one color per vertex.

This is how a 11.1 Mathematica PDF looks like

So the basic problem is that gradients can't be represented in 3D graphics formats?

Because that is what VertexColors are after, linear interpolation right?

@C.E. As far as I can tell, yes. It is the reason, why the PDF size explodes and why they rasterize it. Btw, for me the "AllowRasterization"->False option in PDF export doesn't work anymore.

@C.E. Exactly.

My idea was to do the projection myself and at least remove the hidden faces. But this seems to be a hard task.

@halirutan Does DiscretizeGraphics etc. help at all?

@C.E. Yes, but we loose the gradients and the axes :)

And here you see that we have single-colored faces.

@halirutan Yes, but if the surface is discretized into a set of triangles couldn't one extract the triangles and use a color function to figure out which color each triangle should have, then project the triangles to 2D.

The axes problem I don't know what to do with.

@C.E. I'm sure in the above 2D projection, the hidden faces are still there. The hard part is to cut faces. Let me give a zoom to show what I mean.

@C.E. Look in the upper right corner

The background faces that are "cut" by the foreground faces are the problem.

You would need to check for each polygon if it is completely hidden or cut by a polygon nearer to the camera. Even if you just include the cut polygons you still have to make many tests.

If you include them, you just project from back to front and polygons are drawn over each other to make the correct visual appearence.

ok, I see.

@C.E. If you do

Plot3D[Sin[x + y^2], {x, -3, 3}, {y, -2, 2}] // DiscretizeGraphics

can you export it to a PDF that is a vector graphics? I don't.

@halirutan I haven't tried it, but I guess not. I thought it would be helpful for writing a script that creates a 2D version, but as you say it is thorny to detect which triangles are hidden.

@C.E. Do you know any general 3D vector format that can directly be viewed with a program like Inkscape (svg)?

In a best case, we find out how to reproduce axes, lighting, camera position, etc and export the scene and let another program do the hard stuff.

(I mean we export the 3D scene and get the 2D vector image from a different program).

I see, I don't know any format like that.

I know that PDF and Adobe stuff can include 3D scenes now that are even rotatable inside the pdf.

But somehow this doesn't seem to be open and only available in paid software.

@C.E. If you have Windows/Mac and an Adobe Viewer, you could try this pdf.

I don't have Adobe Viewer, unfortunately.

@halirutan i ended up with the gigabyte aero. 16gb/1060 6gb.. let you know how it goes

@ThomasLackner Sounds nice. I more from the old-school. I really like desktop computers with a decent keyboard and two monitors, but I helped my brother once who wanted a good laptop too. Gaming laptops are the way to go and I guess your machine will do great.

@C.E. Just to show that it is possible:

This is a Mathematica graphics, where I projected the polygons myself. We lose interpolating vertex colors and of course the 3D lighting, that makes some parts darker.

Unfortunately, even this piece of surface is an 8MB big SVG file.

@halirutan yeah me too i just get forced to show up to work sometimes and i would hate to not have it.. cant stand vlc/teamviewer. thanks again

@halirutan that's a very good start, very nice

@Kuba If you restart Mathematica, go to Help -> About, then click System Information, then you get a spinner (Loading...) until the contents appears. Do you know how to create something like this when the contents of the palette will take long to load? Here's an example that does not work.

CreateDocument@PaletteNotebook[
  DynamicModule[{contents =
     ProgressIndicator[Appearance -> "Necklace"]},
   Dynamic[contents],
   Initialization :> (Pause[5]; contents = Framed@Button["Foo"])
   ]
  ]

@Szabolcs I think this is a general way:

CreateDocument@PaletteNotebook[DynamicModule[{init, contents},
   Dynamic[If[TrueQ@init,
     contents,
     ProgressIndicator[Appearance -> "Necklace"]
     ]],
   SynchronousInitialization -> False,
   Initialization :> (
     init = False;
     Pause[5]; contents = Framed@Button["Foo"];
     init = True;
     )]
  , WindowMargins -> Automatic, WindowSize -> 300]

@Szabolcs and here is a duplicate question: mathematica.stackexchange.com/q/59949/5478

@Szabolcs Though this is not what they always do, they use undocumented CachedValue for Dynamic etc. But this is not stable enough, you don't have full control and such Dynamic is triggered too much, that is why e.g. 3D image color function palette is so laggy.

@Szabolcs so I'd use what I've shown above

@Kuba So it's really just the SynchronousInitialization that matters, right? Or is there any reason why it is better to use the init variable instead of a simple assignment to contents?

@e3labs It is not the first time this happened mathematica.meta.stackexchange.com/q/2040/12 Unfortunately, the code detection is not 100% reliable. It still made a large difference in question quality when it was first activated here.

@WReach When using this in a palette, the front end will lose focus in some weird way. I am on OS X, where focus works differently than on Windows. The Mathematica app is still the active one, but no front end window is in focus, and this the palette disappears too (palettes are only visible when Mathematica is in the foreground). Do you have any experience with this? I will need to experiment some more and also try this on Windows.

@yode Try this new TablePaste please, it will work with Excel. I hope you can access this. It might be blocked in China. If not, I'll upload someplace else.

@xslittlegrass @yode BTW, I did learn a little Chinese, but I really can't understand your discussions :) Also, I am not very good at reading characters, unfortunately. This is for when I joined the chatroom where you guys were talking a couple of weeks ago.

@Szabolcs Yep, that is the key. But in case where contents is big cell expression I'd use init or UnsavedVariables :> {contents} to prevent FE from wasting time on that.

Dear friends, education in Wolfram Summer School is free, the cost is only travel and accommodations that Bentley University charges for campus living. Wolfram builds funnest career boosting academic program for 3 weeks in Boston and brings almost 100 top students and professors from around the world.

Networking is amazing. You can be a part of it. Please vote this ad up, apply for the program and spread the word encouraging your friends and colleagues. Feel free to ask me any questions here. Cheers!

@Szabolcs Sorry,I'm in trouble now,perhaps you have to wait me a moment to test that code.

@Szabolcs also, for Dynamic[contents] approach watch out on mathematica.stackexchange.com/q/100828/5478

@Szabolcs it may break even for more complicated content: mathematica.stackexchange.com/questions/136675/…

so the point is to be aware and tweak it :)

It is so frustrating that one must be aware of such arcane details to be able to build something robust and reliable.

@halirutan If we had our own SVG exporter, maybe we could use this to get the vertex colors. Of course, that grows the scope of the project quite a bit. Add ray-tracing for shade and lighting as well and it grows even more...

@C.E. I suspect that a proper graphics exporter can only be implemented directly in the front end. It would have to work with the front end's internal graphics representation (which can take into account things like fonts, glyph sizes, and stuff that depends on it: frames, axes, image padding, etc.) instead of only with Graphics.

When trying to do advanced things, it is a bit frustrating that part of graphics handling is hidden in the front end, and does not seem to be user-accessible or user-programmable

@Szabolcs How to use the .paclet file?

BTW,it is blocked indeed,but I am using APN.So I can access it. :)

@yode Have you seen this?

@Szabolcs You are right, I'm hoping for at best an approximation of what the notebook displays.

@C.E. Thanks.I have now. :)

@Szabolcs Crazy man,your add-on work perfectly on Excel now..

@halirutan I cannot hold my emotion to tell you a excited thing.I suffer a disaster today,my computer cannot start anymore.Then I have reinstalled the Windows System.And now,I can use your SE-Tool.

I'm very exciting to tell you this new. :)

@Szabolcs I just ask,not only localized your this add-on,but I can make a suppose with it to specify.I mean,this case can implement?When all Notebook is minimized,your Palette is float still,and when we click that "Table" button,the data will input into other place(not in Notebook)?Such as our web-browser?

Well,I don't sure whether I have chosen a wrong word form my dictionary,the "Crazy man" is a big praise,even with a little incredible attitude when one can implement great something.I hope it will not result to a unhappyness.

@yode Use PacletInstall with the file as its argument. mathematica.stackexchange.com/q/131101/12 The palette should show up in your Palettes menu. Uninstall with PacletUninstall["TablePaste"].

Thanks,I make it afte Mr. C.E. reminder

@Szabolcs Sorry, I have no experience with this. The behaviour you describe is very strange indeed. There is nothing in the code that (overtly) has anything to do with the GUI or focus. The clipboard function is in the "awt" GUI package, so all I can think of is that there is some strange interaction deep in the bowels of the implementation. Either that, or the the code conflicts with similar logic within the Mathematica front-end implementation somehow.

Hey @Szabolcs, I just noticed this in the MaTeX docs

> Since version 1.6, MaTeX can process a list of expressions with a single run of LaTeX. Structuring your code in a way to allow batch-processing of lists can improve performance significantly. This can be useful e.g. when generating tick labels.

sounds pretty neat

it feels a bit abstract the way that's phrased though

can I suggest adding an example (such as a ticks specification) to that section?

it would make it miles more useful =)

@Emilio, forgot to ping you somehow: have you seen the last update of this answer?

@J.M. no, I hadn't seen it

lemme have a look

@yode This is really nice to hear.

(Of course, he meant your being able to use the SE Tool, not your computer going down.)

haha~

@J.M. looks fine from what I can tell

Wanna see how a ConvNet learns using reinforcement learning and a flavor of fitness proportional selection? See here gph.is/2nSLTzB

I'm still using that code so I'll try and sub in your version before pushing further, to road-test it a bit

now that I look at it, though, SolidHarmonicS[2, 0, x, y, z] produces some funny output

both in your version and mine

(3 z^2)/2 + 1/2 (-x^2 - y^2 - z^2)

6 z^2 - 2 (x^2 + y^2 + z^2) on mine

nothing that a simplify can't solve

but still curious

@EmilioPisanty Yes, I intended for you to go rough on it before finally deciding on whether to use it or not. It looked fine the last time I looked, tho.

@VitaliyKaurov I would love to go, and I even have a project that could benefit, but 3 weeks off is a lot of time...

@J.M. Thanks for the edits. :)

@Szabolcs Yes, this is true. I don't see, why Wolfram doesn't export axes and labels when you use any of the vector formats. Do we know who is in charge of FE and graphics except "the Fultz"

@MartinEnder Well, it's a pretty good question... :)

Thanks :)

@EmilioPisanty Where exactly is that in the docs? There are in fact multiple examples, some in the tutorial and some on the reference page.

You are using the documentation center, not the webpage, right?

@Szabolcs hmmmm, no

szhorvat.net/pelican/latex-typesetting-in-mathematica.html

under performance

third paragraph

don't really know how to get to the documentation center

ok now I do

Open it and search for MaTeX

@EmilioPisanty But there are examples like this even on the website, further up ...

Anyone has an idea how to do this? -->

@Szabolcs hmmmm, so it seems

I guess it's just a discoverability thing

don't get me wrong, it's fine as is

@EmilioPisanty At this point, the documentation center takes precedence. If you have any suggestions like this, do let me know. About the website: I'll only take suggestions on how to make it clearer that people should look in the documentation center. The exception is installation instructions. That remains important on the website.

@Szabolcs I guess that could be made a bit more discoverable from the webpage

@EmilioPisanty Sure, discoverability is important, so keep the comments coming on that. I'll release 1.7.1 with documentation improvements when Mathematica 11.1.1 comes out. I want to wait for that to see if they fix some issues with the documentation styles (before I try fixing them myself).

maybe add a screenshot of the docs center to the webpage, say, just above Usage Examples?

OK.

I know it really sounds selfish, but I think it's a real UX issue

@P.Fonseca yes completely understandable. good thing to consider is how much one can learn in 3 weeks and get useful things done. IN reality 3 weeks flies very fast for the level of projects we do in the Wolfram Summer School.

if I'm looking for some detail of the MaTeX docs, the less mental overhead that the MaTeX search demands, the better

almost by definition, MaTeX is not the main thing on my mind in any real-case scenario

I'll put in this screenshot.

@Szabolcs nice

Hm, that page should have clearer links to the tutorials ...

I thought it had them

have to see what is going on

Also the wording could be made a bit stronger. If indeed the docs is the preferred source for help, there should be some visible text just below the screenshot to send you there

> Note: More usage example are available in the integrated documentation. Search for “MaTeX” in Mathematica’s Documentation Center.

makes it sound like the docs is secondary and that the webpage is primary

OK.

@Szabolcs btw I did end up using MaTeX throughout for my thesis

once you settle into the workflow it did feel like the best option

I'm glad it was useful :-)

tex and mm code, figures and pdfs here if you're interested

so, say, things like this one github.com/episanty/PhD-Thesis/blob/master/6-LES/Figures/… would've been much harder without it

Do you have a final PDF?

You are probably using it more than I am then. If you have ideas about how to make it easier to use, let me know.

@Szabolcs in the root of the repo, or from spiral.imperial.ac.uk/handle/10044/1/43538

@Szabolcs will do. I'm not using as intensively since I submitted (thankfully enough) but I do plan to use it consistently for papers in the future.

You have defended in December. Congratulations!

@Szabolcs naw, that's the official award date, I defended in early October

but same difference though

and thanks!

Do you use SciDraw too?

@Szabolcs initially, but then I dropped it

It is very flexible, but it doesn't allow you not to spend time on details ...

@Szabolcs pretty much. It was just too much overhead.

Fig 3.1 is done with Mathematica too?

@Szabolcs yes

why?

It has a nice hand-drawn feel (of course it is clearly precise enough to show that it's computer made, but it does have that feel)

no, wait

no, it's not

3.3 is

3.1 and 3.2 were done in inkscape if I remember correctly

or possibly actually hand-drawn?

now that you mention it, I've got no idea how I made those

they're from my first paper way back in 2013

man, this is weird.

hmmmm. Probably done on Fireworks, possibly with some pen input.

@EmilioPisanty What version of Mathematica are you using right now?

@Szabolcs right now 10.0 for linux. on the laptop 11.0.1, I think, over windows