I am implementing expectation maximization (EM) in C++ to estimate the parameter of Gaussian mixture model.
The EM is very slow to converge - is there a technique to converge log likelihood in a fast way?
Asked
Active
Viewed 596 times
0
Alex Riley
- 169,130
- 45
- 262
- 238
kahsay kalayu
- 309
- 5
- 16
-
1You should post the code you have now and ask more specific question. – Anton Savin Sep 04 '14 at 09:23
-
In fact, it only converges at the limit of your precision. If you had infinite precision, it would never converge. So choose a different threshold. – Has QUIT--Anony-Mousse Sep 04 '14 at 18:14
-
yes,thank you, i will check by taking other threshold. – kahsay kalayu Sep 05 '14 at 05:48
1 Answers
0
The Armadillo C++ library has a multi-threaded implementation of Expectation Maximization (EM) for Gaussian Mixure Models (GMM). See the gmm_diag class for more information.
mtall
- 3,574
- 15
- 23
-
Hello @Mtall pls can u give me a best document and example on gmm_diag what type of input need? – kahsay kalayu Oct 27 '14 at 11:31