Hi all,

I have written issue
for to attach the first draft of my implementation for LINEST. But there are still a lot of questions. So please have a look at that draft. (Checkmatrix has not essentially changed, but you need this version with correct n and m, if you will test my solution.)

The algorithm for nCase==2 and nCase==3 are nearly equal besides the fact, that the matrices of X and Y values are transposed. How should I handle this? (1) Write all functions in two variants? Then the functions are small and have only one purpose, but the code becomes very long. (2) Introduce a switch between both cases on every place needed. Then the code will become shorter, but the functions become more complex and during execution the very same case distinction is made a lot of times. (3) Transpose the matrices X and Y and then use only one case. That will give short and good maintainable code. But you need double place in memory. How large are those matrices in reality? Is double place in memory possible?

The functions TREND and GROWTH are calculated with the same mathematical solution. Shall I change them in the same way? Then the methods "calculate2" and "calculate3" would be obsolete.

Currently all helper functions are defined as local function and not as methods of ScInterpreter. Do you like to see any of this local functions to become a ScInterpreter method? In the actual implementation the helper functions are all ScInterpreter methods. Why?

What kind of boolean type shall I use? I have now used "bool" locally. But for example in "void ScInterpreter::CalulateRGPRKP(BOOL _bRKP)" a change to "bool_bRKP" would change the declaration. I would have to change it in interpr.hxx too. Does that harm?

Often it has to be tested, whether a matrix was created properly by calling GetNewMat or when pop from stack. Shall I use "if(pMatrix)" or "if(pMatrix.Is())"? Shall I use "errStackOverflow", if GetNewMat does not create the matrix?

QR-decomposition with Householder transformation is a common mathematical method. But it might be necessary to explain the concrete implementation. Please have a look. Shall I omit such comments and attach a separate, commented version to the issue or even write a full text?

My implementation of LINEST uses X-MeansX and Y-MeanY instead of adding a column with 1 to the X values, which you will find in the description in the ODF spec. Does this need a separate documentation?

Kind regards

To unsubscribe, e-mail: dev-unsubscr...@sc.openoffice.org
For additional commands, e-mail: dev-h...@sc.openoffice.org

Reply via email to