Tony Jebara 1998 Here is some fast but ancient code for CEM. This code finds a mixtures of Gaussians to maximize conditional likelihood in a regression setting. To compile type: g++ cemTrain.c++ -o cemTrain To run, type ./cemTrain -help For example, to learn a mixture of 4 Gaussians on the fglass data: ./cemTrain -in fglassTrain.txt -iters 100 -xdim 9 -ydim 1 -complexity 4 -matlab -out fglassModel NOTE: the first time you run the code it will create three lookup tables with a .f extension These will be placed in the local directory. Creating these files is slow but reading them for future will help keep the algorithm running faster. The files in question are: CEMcovLUT.f CEMmuUXLUT.f CEMmuXULUT.f