How to use a remote windows laptop for NetTrain?

Question

I have a windows 10 laptop (with a gpu) at home and a macbook at my office, both with v11.3.

What are steps to use the windows machine remotely for training? That is, I want to launch and run kernels on the windows machine using my macbook's front-end.

I've done this with ubuntu boxes on AWS, but perhaps this is easier/harder?

Answer 1

Ok, here's a file syncing service solution:

Step 1: Get a Dropbox account and make two folders in your dropbox: 'to_run/' and 'finished/'.

Step 2: Paste and run this in a notebook on your Windows box:

Options[run] = {"Print"->False};
(* runs a notebook f and saves its result *)
run[f_, opts:OptionsPattern[]] := Module[
     {nb, pdfName, doneName, tmpName, t, msg},
        tmpName = StringInsert[f, "_IN_PROGRESS",-4]; CopyFile[f, tmpName];
        doneName = StringReplace[tmpName, "_IN_PROGRESS" -> "_DONE"]; t = Now;
        nb = NotebookOpen[tmpName]; NotebookEvaluate[nb, InsertResults->True];
        NotebookSave[nb];
        If[OptionValue[opts,"Print"],
            pdfName = StringTake[doneName,;;-4]<>".pdf";
            NotebookPrint[nb, pdfName];
        ];
        NotebookClose[nb]; CopyFile[tmpName, doneName]; DeleteFile[tmpName];
        msg = StringForm["``: Evaluted `` in ``", Now, FileBaseName@f, Now-t];
        Print[msg]; ChannelSend["example", msg]; (* e.g. monitor things *)
        doneName
    ]

runner[indir_, outdir_] := Module[
    {fns = {}, done = {}, ran, to},
      Print[StringForm["Listening for files in ``", indir]];
      While[True,
        Pause[3]; done = {};
        fns = FileNames["*.nb", indir];
        Do[
            If[!MemberQ[done,f],
                Print[StringForm["Working on ``",f]];
                to = FileNameJoin[{outdir, FileNameTake[f,-1]}];
                CopyFile[f, to]; ran = run[to];
                If[FileExistsQ[to], DeleteFile[f]; DeleteFile[to]];
                AppendTo[done, f];
            ]
        ,{f, fns}]
     ]
]

runner["C:\\Users\\Dropbox\\to_run" , "C:\\Users\\Dropbox\\finished"]

Step 3: Drag a notebook you want to remotely evaluate into the 'to_run/' folder and wait for the notebook result to appear.

Notes:

• This is just a 5-min prototype, you should re-write this to make it robust to notebooks that crash the kernel and handle other issues like name collisions.
• Make sure both laptops have persistant internet access and a dropbox daemon running.
• You can add monitoring easily through a channel with ChannelListener and ChannelSend
• To easily save things that you need in the resulting notebook use DumpSave, PersistantSymbol, or just throw in some Iconize calls.