visualization of latent layer and inducing points

[findoctorlin]

Hey everyone,

Is it possible that within our Gpytorch framework, that we visualize the inducing points and latent layer?

With the DGP regression example in the example folder (https://github.com/cornelliusgp/gpytorch/tree/master/examples/05_Deep_Gaussian_Processes) , which implement the idea of Prof. Deisenroth in the paper "Doubly Stochastic Variational Inference for Deep Gaussian Processes"(https://arxiv.org/abs/1705.08933) I build the model and applied it with a toy dataset and plot the result like this:

Besides this, can we also visualize the mapping from input layer to the latent layer, then the mapping from latent layer to output layer, within our Gpytroch framework?

Thanks!

[michaelcao28]

hi @findoctorlin, have you been able to implement this? I'm trying to visualise a latent GP, where each latent GP has a different kernel structure. Any help would be appreciated!

[gpleiss]

Visualizing inducing points for DeepGP is likely going to be quite difficult. For non-deep GP, see also this issue: #1296 (comment)