Sebastian.p/orth dgp #88
Conversation
|
cleaned up a bit, mypy still complains on quite a few things, but content wise good enough for a review |
…lux into sebastian.p/orth_dgp
|
@SebastianPopescu the PR is now failing on test coverage, mainly due to the untested GPflow customisations. Why do we need to add them here? |
| @@ -291,6 +311,12 @@ def as_prediction_model( | |||
| return model_class(self.inputs, outputs) | |||
|
|
|||
|
|
|||
| class OrthDeepGP(DeepGP): | |||
There was a problem hiding this comment.
@avullo I'm a bit confused, do we really need this class at the end?
I'm waiting for what @sc336 has to say, if he manages to move all this stuff to GPflow in GPflow/GPflow#2027 |
| """ | ||
| lista = [ | ||
| >>>>>>> 53dae0f597e5aa12eea4bc13839a3c94807acd36 | ||
| Cvv(ind_var_u, ind_var_v, k, L_Kuu=l_kuu) | ||
| for ind_var_u, ind_var_v, l_kuu, k in zip( | ||
| inducing_variable_u.inducing_variable_list, | ||
| inducing_variable_v.inducing_variable_list, | ||
| L_Kuu, | ||
| kernel.kernels, | ||
| ) | ||
| ] | ||
| """ |
There was a problem hiding this comment.
What's the purpose of this block of text?
There was a problem hiding this comment.
I think it's a leftover from a previous attempt, can be removed.
Implementation of "Sparse Orthogonal Variational Inference for Gaussian Processes".
Created the following folders:
-- conditionals: needs a specialized form to account for the two different GPs that have to be summed g() and h() as in the paper.
--covariances: needed to compute Cvv and Cvf as in the paper. These covariances rely on the other set of inducing points.
--posteriors.py: different conditionals are needed here.
-- conditionals.util.conditional_GP_maths might have to be re-designed or at least have its name changed to something more sensible.