Probabilistic debugging
May. 21st, 2007 07:43 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
guslTOTAL: Shimizu2006 got 6913/10000 orientations correct
ICA is working too well! Since the error terms are Gaussian, it should perform no better than chance (in this case 50%).
Probabilistic debugging is hell!
Here are some ideas:
* test the Gaussian generator java.util.Random.nextGaussian() ... but a histogram with 3000 points looks really Gaussian, and here my sample size is only 20.
* ...