Apache OpenOffice (AOO) Bugzilla – Issue 91547
ODFF: enhance BETADIST to support probability density function
Last modified: 2013-08-07 15:14:39 UTC
Currently the BETADIST function covers only the cumulative case. In OpenFormula 1.2. the BETADIST function has a fifth parameter to switch between cumulative and density case. OOo should support this parameter too.
reassigning features and enhancements to user requirements@openoffice.org which will be the default owner for those tasks (was introduced some time ago)
Created attachment 55226 [details] necessary changes in header file for new functions
Created attachment 55227 [details] changes for BETADIST
I have attached a patch for the enhancement and for improving the accuracy of BETADIST. This proposed changes give a better accuracy for BETADIST than the current implementation in OOoDEV300m20. But there are still some problems and a final version of the patch will be needed. Please try the patch. You can get the probability density function with an additional sixth parameter false() for densitiy and true() for cumulative. The current form without sixth parameter will still work. For discussion use the mailinglist dev@sc.openoffice.org, please.
Considering for OOo3.1 if remaining issues are solved until then.
Created attachment 55950 [details] changes for new BetaDist
Created attachment 55951 [details] version with ASYM for x near mean and alpha,beta large
Created attachment 55952 [details] Same testcases, the blue values are correct
Created attachment 55953 [details] testcases for the density cases
Hi Eike, this is ready now too. I'll attach the three needed patches. I have made changes in scfuncs.src, interpre.hxx and interpr3.cxx. Please notice first to implement the patch of i91602 to get expm1 and log1p. In addition I have added columns to the test document containing the values that the current OOo gives. So you can see, that even in those cases, where we are less accurate than Gnumeric and CAS, the improvement compared to the current solution is so considerable, that you should not hesitate to apply the patch.
Created attachment 58481 [details] test document to compare new and old values
Created attachment 58482 [details] patch for source/ui/src/scfuncs.src based on odff05
Created attachment 58483 [details] patch for source/core/inc/interpre.hxx based on odff05
Created attachment 58484 [details] patch for souce/core/tool/interpr3.cxx based on odff05
In cws odff05: revision 264879 sc/source/core/inc/interpre.hxx sc/source/core/tool/interpr3.cxx sc/source/ui/src/scfuncs.src This yields indeed a _much_ better precision than before. I noticed that for unxlngi6 non-pro in the test case documents the calculated values slightly differ from the ones copied to the "OOo" column. This may be due to different compilers, optimizations et al. Interestingly in most cases the calculated values were a little bit closer to the MuPad values than the copied values, but only in the second-last displayed digit or so. I'll recheck in a product build.
Reassigning to QA for verification.
Created attachment 58896 [details] Testdocuments for Test Case Specification
Created attachment 58899 [details] TestCaseSpecification
verified in internal build cws_odff05
Verified in OOo310m9 on WinXP and Fedora9 Closing Li Meiying