DQN Cartpole Example Does Not Learn

[ss32]

Opening because #683 is closed and the issue persists. When running the vanilla code from the Q learning example, even for 10k episodes, the agent does not learn. The reward is constant at 1.0, and losses increase over time.

$ python3 -c 'import torch;print(torch.__version__)'
1.10.0a0

cc @vmoens @nairbv

[svekars]

Closing this since it seems to be resolved. Feel free to reopen if not.