05 May 2017 8 8K Report

Hi,

I am trying to fit Gaussian process to learn a distribution from input sequence to output sequence. I am relatively new and have several questions:

1) For time series prediction problems, if you have a sequential input you can feed them without stacking inputs into a vector while using RNNs or LSTMs. Do Gaussian processes have such a structure? If no, is it still make sense to use GPs to learn output sequence from input sequence?

2) When I try to fit GP to training data with 400 samples and 18 features it has really difficult time. Is there a way to batch the data and use the learned parameters from one batch to another as in neural networks.

Thanks in advance.

More Kko Kko's questions See All
Similar questions and discussions