lawrennd
commented
8 years ago Hi finmod,
Yes, we're aware this needs improvement, it's on our to do list, but not yet in the pipeline. I don't have specific answers to your questions, @jameshensman might be able to help and @nfusi has made the airline delay data available in the past, but both are very busy. If you do manage to disentangle some of it then do please feel free to do a pull requst, we'd love any contributions!
Neil
On Tue, Feb 16, 2016 at 4:53 PM, finmod notifications@github.com wrote:
There is only one notebook example, SVI.ipynb, of stochastic variational inference. In this example, X = input, Y = multiple outputs (2 columns Y1 and Y2) and Z = inducing variables.
Z = np. random.rand(20,1) where 20 appears to be the batchsize x the number of outputs. Is this correct?
Further down the number of kern.RBF(1) and kern.White(1) appears to be full rank output matrix (2). Is this a necessary condition or low rank approximation can be investigated this way? e.g. an output matrix of rank 20 can be approximated with RBF(5) and White(2)?
Better docs or tutorial notebook(like UK apartment price or airline delays) for GPy.core.SVGP would help because this is really cryptic.
— Reply to this email directly or view it on GitHub https://github.com/SheffieldML/GPy/issues/303.
BrunoQin
There is only one notebook example, SVI.ipynb, of stochastic variational inference. In this example, X = input, Y = multiple outputs (2 columns Y1 and Y2) and Z = inducing variables.
Z = np. random.rand(20,1) where 20 appears to be the batchsize x the number of outputs. Is this correct?
Further down the number of kern.RBF(1) and kern.White(1) appears to be full rank output matrix (2). Is this a necessary condition or low rank approximation can be investigated this way? e.g. an output matrix of rank 20 can be approximated with RBF(5) and White(2)?
Better docs or tutorial notebook(like UK apartment price or airline delays) for GPy.core.SVGP would help because this is really cryptic.