Increasing lsqcurvefit/nlinfit speed

Hi,

I'm looking to increase the speed with which lsqcurvefit is running in my code. The "fitting function" attempts to find a solution by using a lookup table, who's axes are some of my free paramters. I suspect speed is taking a hit because my table is almost 3GB. Right now, I'm sending the table to the anonymous function instead of loading it within the function each iteration (see below - "crossSecSimC2H4" is the database).

[estimCustomCSec(i,:),resnorm,residCsec,~,~,~,Jac] = lsqcurvefit(@(freeParamsC,waveExperiment)fitCustomCrossSecData...

(freeParamsC,waveExperiment,numDens(i),L,absLaser3(:,i,1),bCoeffC2H4,bCoeffCH4,xC2H4range,xCH4range,crossSecSimC2H4...

,crossSecSimCH4,waveExperiment,waveDatabase),freeParamsC,waveExperiment,absLaser3(:,i,1)',[],[],options);

Am I losing speed because even "sending the file" to the function takes long? Or is there no copying of data, in which case the fitting process itself is computationally expensive?

Answer 1

[Matt J]

It would definitely be faster to attach the table as an external variable to an anonymous or nested function, than to freshly load it from disk every iteration.

The slow-down might be coming from the time required for table-lookup. A direct computation can sometimes be done entirely within the CPU and its cache, whereas a lookup process, especially in a table that size, would require frequent RAM access.

[Raghav Poddar]

I see, that makes sense. Thanks!

[Matt J]

You're welcome, but if your question is resolved, please Accept-click the answer.