EOF and Segmentation Error after upgrading to python=3.8 and replacing baselines with stable_baselines

[lucasosouza]

Not clear if the error is from replacing python, baselines or one of the previous PR as it wasn't tested in between. Might also be related to changing all arrays to torch tensors.

@vkakerbeck please add more information if you have them

[lucasosouza]

PR #35 fixes reference to Ray Tune, so it can be installed with an environment with python 3.7

[vkakerbeck]

I tried running the code from before the switch to stable-baselines3 (commit a561eed) and the EOF error shows up there as well. Also in the newest commit with python 3.7 doesn't fix the problem.
As you said, it seems to be a memory leak and the EOF error then sows up after some time once we run out of memory which is why we didn't notice it in the short test runs.
I think it most likely comes from changing all the arrays to torch or something else we did around that time. The memory leak was not there yet when I tested the original implementation of differentiable disagreement (commit 16c61b0)