@Zviovich Yes, I can replicate it.

@same error then?

@C.E. I reported this error to WRI but they state they can't replicate. Please if you can document in an answer you see the same issue in your system, I can point to production support that this happens to other users too

@Zviovich This is sort of interesting: URLExecute works, but URLExecute[url, <|"Cookies" -> {}|>] doesn't work. In my $Cookies I have cookies related to the channel broker which will be sent along. But the browser or the terminal don't have that cookie. I can type this up into an answer as you suggested, so you have something to point to.

@Zviovich Sorry, I made a syntax mistake. The answer will have to be more limited to just replicating the problem.

What's going on here?
eq = y''[x] == y[x]^2 - x;
ibcs = {y[0] == 0, y[N1] == Sqrt[x]};

N1 = 3;

sol = NDSolve[Join[{eq}, ibcs], y[x], {x, 0, N1},
  Method -> {"Shooting",
    "StartingInitialConditions" -> {y[0.0] == 0, y'[0.0] == 0}}]

@MMM You initial condition should read Sqrt[N1]

@halirutan Why is that?

@Szabolcs are you around?

@MMM You are specifying an initial condition y[x0]==f[x0] and on the left side you put in the specific value for x0 and on the right side you don't. That just doesn't work.

@halirutan Got it. Thanks as always. BTW, for N1=15 the solution blow up/numerical noise.

@Szabolcs thanks for tips. I yep am aware what git archive does. I'm planning to compress tags. It would be nice to have nice sync between latest tag and paclet version

@Kuba One difficulty is that sooner or later you might feel a need to "build" the package from sources, i.e. do something extra before the final version is produced. In that case the git report would hold the sources only.

@Kuba I took the freedom to fix some typos in your moderator post.

@halirutan that doesn't sound good ;P but thanks ;)

ah, there are nominations now? let me see

@Szabolcs yep you are right, I'm already building it but commiting it anyway as it was allowing me to manage ceratin things easily. But right, in general this will be a problem

@Szabolcs if you nominate yourself I will resign as I think you fit there better

Currently I have a mess where I do build, yet I still do commit some of the built files to the repo. I commit the LibraryLink binaries because GitHub makes it easy for me to move them between the different computers/platforms where I build them. I really don't know what's the best way

@Szabolcs yep I'm also confused. I keep diffable part + very crucial files as git repo and the rest in analogous subdirectory in my google drive.

I need to run, meeting in 2 min ... will be back after lunch

Which allows me to switch between different machines as I use 3 on daily basis.

@Szabolcs Thanks for tips, see you

@Kuba The git-lfs support makes it really easy to include large binary files that shouldn't be diffed.

@halirutan good to know, will try this eventually, but now I'd have to setup it for private repository server if I don't want to use github, right?

@Kuba If you don't want to pay for GitHub like I do, you could go to Bitbucket. Free private repos.

@halirutan the setup is quite complex, I play with github but also have a repository on a private server.

Anyway, I will keep that tip in mind and if I find more time I will test it. Now I have enough things to experiment with :)

@Kuba I'm not sure how git lfs works on a private server because GitHub hosts the binary files on Amazon S3.

@Szabolcs fyi, assuming you are in package dev directory this will produce valid .paclet from a subdirectory Package:

from a commit associated with 2.1.2 tag

2.1.2. from .paclet doesn't mater at all but 2.1.2 from the prefix matters.

if you have prefix=Package/ then PacletInstall will extract it to Repositories / Package

and everything is fine, appropriete files will be overwritten but if any of them is locked mma will freeze

Is there any overall difference between using Switch vs Which for something like this,

I feel like the Which is more readable, but in general I know the pattern matcher is cheaper than invoking the evaluator.

but maybe in this case I'm evaluating anyway with the pattern test

@JasonB I've never looked into performance differences, so I just go with readability. I use the same rule of thumb I use for Select vs Cases. If the criteria is more logical then I use Select/Which, if it's more structural/atomic then I use Switch/Cases.

@MichaelHale Thanks, that makes sense. going with Which in this case I think

Playing with historical stock index data... Predict is able to produce considerably better one-year forecasts on index development than just mean growth and standard deviation... :o

I suspect I shouldn't rely on this though.

@kirma I'd put my whole IRA in Mathematica's hands, What could go wrong? :-D

@JasonB "IRA" ?

individual retirement account

I put money in every year, and can shuffle around the distribution to different stocks, bonds, funds

Oh well. I think there's no such thing around here.

then, if I had put in after-tax money, when I pull it out the earnings on it are tax free

@JasonB You live an adventurous life. :P

@kirma maybe you live in a country with nice social welfare for the retired, sigh...

I do need to decide investment strategies for a sum that, at least most western countries, is more than life savings of an average person... :o

Sure, primary solution is to just diversify. While I don't really believe on my personal modelling skills, surprisingly long-term macroeconomic investment simulations offered by local private bankers correlate quite well with my Mma back-of-the-envelope risk models.

Which probably means everybody is equally blind...

Frankly, I think Predict is likely to memorize the dataset in some non-trivial way, and there is just too little training data to build a general model on these timescales.

I use NDSolve to solve a transient heat equation in polar coordinates. For this I need a PeriodicBoundaryCondition equivalent to u[r,phi]==u[r,phi+2 pi]). But PeriodicBoundaryCondition has a notion of Source Boundary and Target Boundary. The results are not the same in the 2 cases. Furthermore any linear combination of these 2 solutions are also a solution. Nevertheless my problem has physically only one solution. What I'm missing ?

Note : The conservation of the flux at the boundaries is respected in both solutions.

Note 2 : if I had the wave equation instead of the heat equation this would be normal.