Open karlnapf opened 10 years ago
@karlnapf I have set a constant number of points in https://github.com/shogun-toolbox/shogun-demo/pull/50 instead of depending on training points, please check if thats better.
seems like a good idea, maybe we could even have an option for the number of testing points? might also help for the predictive distribution
i like the new grid!
there are lots of errors if you switch between fitc and exact inference with and without learning hyperparameters, could you play a bit and investigate ?
Hmm.. I think we need to reset default parameters before learning, will do that.
@Saurabh7
the tube should have a finer grid. that is, the test data for the tube should have the same resolution as the plotted mean prediciton.
let me know if you have questions