stober / gmm Goto Github PK

Hi! I'm planing to use your lib for for nan filling. Your em algo gives inconsistent results though. If I generate np.random.multivariate_normal([0,0,0],[[1,0.8,0],[0.8,1,0],[0,0.8,1]],4000)

and choose 1 component, I get
[[ 1.05881648 0.71821905 -0.19137837]
[ 0.71821905 1.18875572 0.40823723]
[-0.19137837 0.40823723 0.92021874]]
as sigma estimation. Even manually setting the true cov matrix doesn't help, so it is not a local optima problem. I think I'll try to fix it

But I see no obvious mistakes in EM or density estimation

hey man, I find a problem in em process, you use 'response / np.sum(response, axis=0)'. the value of np.sum(response, axis=0) may be zero, so that the final result is nan.

When trying to install gmm, , the following error appears:

Error: package or namespace load failed for ‘gmm’ in dyn.load(file, DLLpath = DLLpath, ...):
unable to load shared object '/Library/Frameworks/R.framework/Versions/4.0/Resources/library/gmm/libs/gmm.so':
dlopen(/Library/Frameworks/R.framework/Versions/4.0/Resources/library/gmm/libs/gmm.so, 6): Library not loaded:      
/usr/local/gfortran/lib/libgomp.1.dylib
Referenced from: /Library/Frameworks/R.framework/Versions/4.0/Resources/library/gmm/libs/gmm.so
Reason: image not found

We tried other ways to install the package, without success:

• We tried to uninstall and reinstall R and RStudio
• We tried older versions of R (3.3.3 and 3.3.6)
• We tried to copy the files directly on the R library (/Library/Frameworks/R.framework/Resource/library),
• We tried to install the package as a binary (install.packages("gmm", type = "binary"))

Could you please give us a hand solving this problem ?

Thanks in advance.

there is a singular matrix when processing my data