Well this is a new bug...

I quit Mathematica forcibly but its windows aren't actually disappearing

Another bug that I've reported to support but forgot to mention here. If you try to move the selection in an empty notebook it hangs.

e.g.:

CreateDocument[]

(* in a new cell *)

SelectionMove[nb, All, Notebook]

Ah sorry that one's insufficient. You need two things at once I think:

MathLink`CallFrontEnd@{
  FrontEnd`SelectionMove[nb, All, Notebook], FrontEnd`NotebookWrite[nb, {}, None, AutoScroll -> False]
  }

@b3m2a1 It's not that FE that makes it so big.

@ChrisK I think what Arnoud meant was that editing notebook is disabled. So one can't use notebooks normally (apart from viewing them).

Can we add more options to "Close - This question belongs on another site in the Stack Exchange network"?

Specifically, stats.stackexchange.com would be nice on occasion

@b3m2a1 You are on Mac, right? On Windows there is no problem with hanging.

Does ParametricNDSolve do anything more than cache the previous results for the exact same parameters?

In terms of speeding up repeatedly evaluating the same ODE with different parameters

@b3m2a1 I'm really curious if this would enable running "hobbyist" neural networks code on EC2 instances in a way that wouldn't be a breach of license and interact gracefully with Home edition license on my laptop...

Thankfully FAQ would indicate there are no restrictions on using it on a cloud platform. I suppose it can't be really run as a remote kernel, though, but packaging neural network training workloads from the front-end, deploying on the cloud, running as WLS script and fetching back doesn't sound awfully impractical either. Main problem there is that transfer of large datasets out from, say, EC2 costs money.

Just hoping they don't put arbitrary restrictions on usage of CUDA, or such.

I could imagine a relatively simple system for deploying arbitrary batch jobs on the cloud and fetching results back when they are ready. Or do such solutions already exist?

@L.K. Have you seen AxesEdge?

@C.E. That's great news. But after development (and if an open source application is blessed by Wolfram), how would a user of the software get the engine? It seems right now the user would need to download the engine and "pretend" to be a developer.

For me if the engine was put into the R library and maintained by Wolfram, that would be great. But maybe that's too much to hope for.

@RolfMertig Interesting. Even if you make a new document with nb = CreateDocument[] and then in a separate evaluation do MathLink`CallFrontEnd@{ FrontEnd`SelectionMove[nb, All, Notebook], FrontEnd`NotebookWrite[nb, {}] }?

That hangs for me

Hmmh. I'd assume Channel* functions using Wolfram Cloud broker cost money, or credits in a manner or another. Where would that pricing be available?

Documentation implies that other brokers would also be possible, but I can't find a single example of such a case.

@kirma I don't think they do... You could see if using them eats up $CloudCreditsAvailable.

@b3m2a1 Then again, I was just thinking it could be used as a sort of connectionless replacement for WSTP on own host. A middleman would make it impractical for large amounts of data.

Ah interesting. You could probably dig into the source to see if there's any way you can adapt it to work with a local configuration:

PacletFind["ChannelFramework"][[1]]["Location"]

@b3m2a1 Then again, WSTP over SSH with retries would probably be fine (I'm thinking of building stuff on Wolfram Engine in dynamic EC2 instances + S3...)

Just that HTTPS would make more sense as a transport which is more naturally tolerant to interruptions...

@kirma I'm interested in this as well, mainly for cloud training of neural networks on GPU-enabled servers (eg scaleway's $1/h GPU servers). I'm fairly confident they're working on built-in cloud training for NNs, but I don't expect to see it for another couple of years

@CarlLange I'm mostly thinking of AWS because that would make any code written for the task more likely to be reused... but of course one should try to generalise any such code.

@kirma Yeah, I mean, it should all basically be the same

Many cloud providers have S3-API compatible object stores as well

Clarification to my last message: "I'm fairly confident WRI is working on built-in cloud training for NNs in WL, but I don't expect to see it for another couple of years"

This far I've figured out a practical solution would include VM instantiation, kernel-to-kernel communication and persistent storage. I haven't really looked on other platforms how it would be done...

NNs being just the most obvious low-hanging fruit.

If you feel like it, feel free to post an answer: mathematica.stackexchange.com/questions/181115/…

I'm just somewhat concerned that if WRI license for the Wolfram Engine is as reasonable as one can optimistically think at the moment, this would still seem like a license revenue threatening usage from their perspective. At the same time it would make Mma much more attractive as an experimentation platform, IMO.

@CarlLange I'm waiting for further information (blog post) on WE, as was suggested earlier on chat. Not to mention it probably takes weeks to hack anything convincing together to be convinced of it... ;)

Yeah, I don't really know what they are planning wrt licensing. Sometimes I'd love to learn about their revenue flow, because I haven't got a clue where the money really comes from, and whether potentially losing a number of "Professional" licenses to WE would actually make a difference to them.

I suspect there are a lot of people that currently defy their license, so perhaps it won't make a difference

@kirma No worries, I agree.

It should be a license violation to use free WE for any commercially valuable activity. It's just that they might find out that they have made it "too easy." Just wondering about it.

@Szabolcs I guess I don't know exactly where the size of all this comes from. There's a WolframPlayer app bundled directly into the WolframEngine app which copies the entire Mathematica layout except the documentation, basically. I wonder how much of what's in there is truly necessary. It would be nice to have a much more light-weight engine for quicker deployments. Maybe even a front-end-less one, as that'd save a full GB of space.

(based on looking inside the directory and computing the size of the FrontEnd system files)

Wait no it's only ~100MB. Nevermind it's not that simple :|

Oh wild they package a full Java install in Mathematica

I didn't realize there was any environment without a working Java runtime

@b3m2a1 The problem is the Java version, not that there is no Java available.

I see. If you have an up-to-date Java runtime is it safe to delete that folder?

I guess it's not a huge amount of data in the long-run

But my computer is getting old and overstuffed and I don't want to delete most of my archived files :)

@b3m2a1 Well, if up-to-date means Java 11 or 9, then I wouldn't risk it.

Long-run it's only like 200MB I guess...

I'll prune it out of my old Mathematica installs though.

The historical value of having an 11.0 to test against is rapidly declining

@Kuba since you asked that question about making atomic structures, here's a really simple one I just made to simplify my work:

getFD[fitData[a_], k_] :=
  Lookup[a, k, a["Fit"][k]];
RegisterInterface[fitData,
  {"Fit", "Points"},
  "AccessorFunctions" -> <|
    "Keys" -> getFD
    |>
  ];

Does it do anything fancy? Of course not, but all the boiler plate is handled automatically, so little reason not to use it instead of a plain Association

Where I'm using it in practice is like:

fitData@<|
  "Fit" ->
   LinearModelFit[
    fff,
    Flatten[{\[FormalR]1^Range[4], \[FormalR]2^Range[4]}],
    {\[FormalR]1, \[FormalR]2}
    ],
  "Points" -> fff
  |>

And then I have a fitData object I can work with quickly and easily

@MichaelE2 Thanks. That solved my problem finally. Thanks to CarlWoll and klgr

@kirma There is a Terms of Use here: wolfram.com/legal/terms/wolfram-engine.html

Is anyone else having issues with copy/paste in Mathematica? It sometimes only pastes part of the copied stream if you paste into a plaintext env:

CloudDeploy[
 Notebook[
  {
   Cell[
    BoxData[
     ToBoxes[
      Column[
       {
        Text[
         Style[
          "Find the solution to the following system of linear equations:",
          FontFamily -> Times,
          FontSize -> Large
         ]
        ],
        TraditionalForm[
         Style[
          Equal[
           Plus[
            x,
            3 * y
           ],
           7
          ],
          FontSize -> Large,
          FontFamily -> Times
         ]
        ],

Like try copying that out of a notebook

I only get this:

s_will_be_dead_in_60_days_because_WRI_did_us_dirty\",
 Permissions -> \"Public\"
]"

If I copy as InputText rather than PlainText

@ArnoudBuzing any updates on the paclet repository status?

It'd take more time and energy to make either my or @C.E.'s servers successful than I think either of us have