New episodes cannot be generated

[Raphaelzrf]

After training for a period of time, new episodes cannot be generated.The phenomenon manifested is that training stops. This bug is hard to reproduce because it always appears after a long period of training, such as 100 epochs. I suspect it may be related to the communication between multiple processes.

[Raphaelzrf]

I saw the same issue mentioned in #125 and the solution was to restart training, but I would like to know the cause of this behavior.