munskygroup / ssit Goto Github PK

As MCMC can be a pain to tune for efficient mixing, and as the current usage of MH in my Bursting Gene example is incredibly slow to obtain high ESS, a fair comparison between the MatLab SSIT implementation and a Stan implementation that uses NUTS to perform HMC would at least use FIM for proposals as per Munsky Group protocol. Stan's machinery is already such that the user need not necessarily spend ages autotuning or create their own operators to improve mixing @alexpopinga

Line 904 of RonModelPipelineJuly11_2024.m (terminalVersionDevelopment) specifies the priorCoVariance for the evaluateExperiment function called from SSIT.m to compute the FIM, leading to the following error:

"Arrays have incompatible sizes for this operation.

Error in SSIT/evaluateExperiment (line 1198)
fimTotal{is} = fimTotal{is} + cellCounts(it)*fimResults{it,is};

Error in RonModelPipelineJuly11_2024 (line 903)
fimTotal = ModelGRDusp100nM.evaluateExperiment(fimResults,ModelGRDusp100nM.dataSet.nCells,..."

Runs with an empty array (default).

After completion of testing and debugging, refactorise.

MH is currently embedded inside the computeLikelihood function of SSIT.m