Anthony Corso

Allow the reward function to accept heuristics from the simulator natively (these are quantities that guide the search algorithm like how the close the vehicle was to a collision which)

The package was originally designed to work with a handful of POMDPs from the POMDPs.jl ecosystem. Now that it is more general, we should probably change the design so that...

Is it possible to bump the compatibility from GPUArrays 8 to 10? This should allow for compability with the most recent versions of CUDA.jl

There seems to be a single terminal = true flag in each of the half cheetah datasets. Do you know why this is the case? The half cheetah gym environment...

[Bug] Broken example `Exact GP Regression with Multiple GPU`

2

# 🐛 Bug The Multi-Device example "Exact GP Regression with Multiple GPU" is currently producing a runtime error on a standard 8-v100 node. The error is `Expected all tensors to...

# 🐛 Bug Multitask GPs (using the LMCVariationalStrategy and MeanFieldVariationalDistribution) are seemingly incompatible with the NNVariationalStrategy. This makes it difficult to train multitask GPs with very large numbers of inducing...

[Bug] _diagonal() fails when covariance is computed using keops kernels

2

# 🐛 Bug Calls to _diagnonal() error when keops kernels are used. This showed up when tring to use keops kernels in a Multitask/Multi-output approximate GP with SVI. But below...

[Bug] Fantasy Models for Multitask GPs are broken

5

# 🐛 Bug Getting a fantasy model for a simple multi-task GP throws an error ## To reproduce Here is a minimum working example of the bug ```python import torch...