Fix and/or clarify data filename parameter of formPropensitiesGeneral

[dmitrisvetlov]

Some of the problems can be seen by running the "Designer specification" section of DemoScript.m in the SequentialExperimentDesign branch. There is a dataFilename parameter to the formPropensitiesGeneral method; since the code doesn't document any constraints on this name, I have been using trial-and-error to figure out how best to set this parameter for the SED codes (particularly in the model factory).

Here is the relevant code:

sed = SequentialExperimentDesigner();
sed.DataType = ExperimentalDataType.Empirical;
[~, dataFilename, ~] = fileparts(tempname(pwd));
addpath(genpath('..'));
sed.Model = DiscoverableModelFactory.createModel("GR", dataFilename);

Here is the MATLAB output:

Error using sym/matlabFunction>getOptions (line 812)
The value of 'File' is invalid. It must satisfy the function: isFile.

Error in sym/matlabFunction (line 154)
opts = getOptions(args);
       ^^^^^^^^^^^^^^^^
Error in ssit.Propensity>sym2mFun (line 850)
            exprHandle = matlabFunction(symbolicExpression,'Vars',{states,parameters},'File',fn,'Sparse',false);
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error in ssit.Propensity.createAsHybridVec (line 401)
                            sym2mFun(signHybridFactor*expr_x, false, true, nonXTpars(:,1), speciesStoch, [], false, true, prefixNameLocal);
                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error in SSIT/formPropensitiesGeneral (line 157)
                        ssit.Propensity.createAsHybridVec(sm, obj.stoichiometry,...
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error in DiscoverableModelFactory.createModel (line 119)
                    m = m.formPropensitiesGeneral(append(dataFilename,"_S",num2str(ind)),true);
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Note that I had to make several changes first.

1. The original line 119 in DiscoverableModelFactory.m utilized array concatenation ([dataFilename,'_S',num2str(ind)]) to create a scalar prefixName. For whatever reason, the elements weren't being concatenated successfully, so I used the append method instead.
2. Even after that fix, an analogous problem (a non-scalar argument) arose on line 222 in Propensity.m, so I also changed that code to use append (see this commit).
3. In (randomly) generating my temporary filename, I am also stripping out the path and extension so that no path delimiters are present.

My first guess, reading the error message, is that either no file has been created with the exact filename provided in the call to matlabFunction, or some incorrect value of fn is being passed.

[Munsky]

Hi @dmitrisvetlov
I couldn't reproduce the bug. I did have to change the filesep to '/' to get it to work on my mac, but after that everything ran and the model solution looks like it is working.

Is it possible that it is a filesep problem in Windows?

[dmitrisvetlov]

Hi @Munsky,

Thanks; I am able to load data and solve the model using the FSP using the forward slash to finalize the path to the data file. That said, on Windows, the pwd command does return a filepath that uses the backward slash, consistent with nomenclature on Windows, so [pwd someFileName] does result in a string with a mixture of separators if the forward slash is used. So I'm not sure why using the Unix-style separator fixes the problem. If anything, it should introduce one: a mixture of separators where previously only a single one was used.

Regardless, I am still unable to use the "data filename" I randomly generated when calling formPropensitiesGeneral. Before loading the data, I am only able to create the model in the first place if I

• Omit the call to formPropensitiesGeneral in the factory method OR
• Use some other string as the filename, e.g. GRTest.

I considered the possibility that the problem lay in the fact that the filename I'm generating is returned by tempname() as a char array, whereas the other elements are strings. But even if I intervene and cast dataFilename to a string, the problem of a non-scalar prefix in createAsHybridVec still occurs.

We can sit down together and troubleshoot this when we next meet if you can't reproduce this in the interim.