search

pmelchior / pygmmis

Gaussian mixture model for incomplete (missing or truncated) and noisy data
MIT License
98 stars 22 forks source link

Background amplitude fitting as part of normal EM, not hierarchical #4

Closed pmelchior closed 6 years ago

pmelchior commented 6 years ago

Currently, we make a distinction samples belonging to background vs signal, and only the signal samples are fed to the GMM. This is suboptimal for samples that have comparable probability of being from background or the signal. A more natural option is to compute the q_ik for the GMM and the background, and then to renormalize the amplitudes of background and signal. This could reduce some of the code complexity and give better results for the GMM means and covariances.

pmelchior commented 6 years ago

Done with 425bf62a51225ac803f3a4c595b1ddab735dfbcb