-
At the moment we use lbfgs_b from scipy as the default optimizer in GPflow.
Particularly with #185 and using float32 and float64 we have some anecdotal evidence that lbfgs_b is aggressively finding …
-
Is there any interest in adding a quasi-Newton based optimizer? I was thinking of porting over:
https://github.com/tensorflow/probability/blob/master/tensorflow_probability/python/optimizer/bfgs.py…
-
### 🚀 The feature, motivation and pitch
Despite not being explicitly documented, [`torch.triangular_solve`](https://pytorch.org/docs/stable/generated/torch.triangular_solve.html) appears to support…
-
It might be a stupid idea, didn't think it through from code perspective, but couldn't Ngboost predict the Normalizing Flows' parameters as well, so the final outcome distribution could be transformed…
-
I'm interested to implement the Neighbourhood Algorithm (NA) in Python. This algorithm is quite popular in the geophysical community for solving optimization/inversion problems. It is detailed in [thi…
-
@MikeInnes and I had a long discussion of this.
One might reasonably want to use a different optimizer for different parts of the network.
Or equivelently, different optimizer params (e.g. learni…
-
Hi, I had read the paper "Dont unroll adjoint", and from my understanding, zygote had invented a new AD algorithm. However, there is also lots of AD compiler out there, and is there any head to head b…
-
This issue carries on the discussion at https://github.com/JuliaStats/Roadmap.jl/issues/14 with the intend to make things more concrete. To start with this, I see a list of apparent questions and TODO…
-
**Describe the bug**
Thanks a lot for great and useful library.
I found that NormalDistributionsTransform::align may not work correctly by gimbal lock.
It's depend on user specified "guess" a…
-
**Is your feature request related to a problem? Please describe.**
Currently we operate on flat `[0,1]^D` representation, however this is artificially imposed for convienience and requires ravelling …