Evaluate notebook from other notebook

Question

I have a notebook ("notebook 1") which in the beginning sets some parameters, does something depending on their value and saves the result to a file, say:

p=5;
result=p^2;
result>>"result";

I want to run this for many combinations of parameter values, so I'd like to automate the process. The notebook is large and I don't want to pack it into a single function of the parameters, as that would reduce readability.

So I want to run it from another notebook ("notebook 2"), effectively as if it were a function. I also want each run to be on a fresh kernel.

This is partly done by the following code

runWith[pval_]:=(
p=pval;
Protect[p];
NotebookEvaluate[notebookaddress];
Unprotect[p];
ClearAll[p])

where notebook 1 uses a parameter named p and this code runs it with value pval assigned to it. The protect is necessary because the notebook tries to set it to another value, as in the example above. However this keeps the kernel on successive runs.

Adding Quit[] to notebook 1 doesn't work, as it is run in the same kernel, so this will quit the kernel of notebook 2 as well.

A solution would be to let notebook 1 evaluate in another kernel, and then quit that kernel, both from notebook 2, but I can't find if this is possible.

Answer 1

The best way is to use bash script!

First you need to save your script as .m file and "save as" option, wont work alone! You have to convert it to "code" format, by select all cmd+8 in OS X if I'm not wrong or select all and then fromat->style->code.

Second, we tell mathematica to run its kernel through a command. First we give the path to the Mathematica to run the kernel, which will run your script! Your code will look like below, the part which says result=p^2 sends the results value to the .m script which will be evaluated in the kernel.

SetDirectory["/Users/ray/Desktop"];
p = 4;
Run[Evaluate[
  "/Applications/Mathematica.app/Contents/MacOS/MathKernel -run \
results=\"" , p^2, "\"", "-script test.m > itest.txt"]]

your results will be saved in itest.txt you can easily open that in mathematica and process the results. You need complete path to .m files by the way, pay attention to setdirectory!

Answer 2

The best solution I have been able to come up with so far has the following properties:

• Nothing needs to be modified in the 'worker' notebook (the one that's run many times), so it can be run manually as well without any changes.
• The 'distributing' notebook, the one telling the 'worker' to do the runs, can override value assignments in the worker notebook.
• Each run of the worker starts with a clean slate, previous runs can't interfere with the current one.
• Any printing in the worker notebook is suppressed, so the distributing notebook is not flooded with prints.
• Error messages still show though, and so do any Monitor commands.
• Saving is done inside the worker notebook, with Put, making it easy to access.

The solution is as follows,

runWith[pValue_Integer] := (
Begin["WorkerContext`"];
p = pValue;
Protect[p];
timeTaken = 
Block[{System`Print = dontPrint, 
  System`StylePrint = StyleDontPrint}, 
 Quiet[NotebookEvaluate[NotebookDirectory[] <> "worker.nb"],Set::wrsym]
    //Timing // First];
Print["p = "<>ToString[p]<>" completed in " <> ToString[timeTaken] <> " s."];
Unprotect[p];
Unprotect["WorkerContext`*"];
ClearAll["WorkerContext`*"];
Remove["WorkerContext`*"];
End[];);

So starting with a clean slate is done by working in a new context, and clearing that context after each run. Overriding value assignments is done just by protecting them (I suppress the warning message this generates with Quiet), printing is suppressed by just redefining Print in a Block.

The 'worker' notebook would be as in my question (except that it would save to "results"<>ToString[p]).

One functionality which this does not have is to be easily parallelizable. If I run ParallelTable[runWith[p],{p,1,10}] it complains with FrontEndObject::notavail: A front end is not available; certain operations require a front end.

If anyone has a solution which has all of these properties and is also able to run in ParallelTable, and is not significantly more complicated, I will accept their answer.