search

eleurent / rl-agents

Implementations of Reinforcement Learning and Planning algorithms
MIT License
591 stars 153 forks source link

Logging Error #62

Closed skynox03 closed 3 years ago

skynox03 commented 3 years ago

Hi,

I am facing a problem with logging file of the results after training. Whenever I run a training by loading a saved model, the output log file (containing the reward details of each episode) of the old training is over written by the the results saved in the new log file. for example, i ran a training for 3000 episodes, then a new training of 5000 episodes. But now the log file from 3000 episode folder has the results from 5000 episode training saved in it. I have attached a screenshot too, the results are identical. How can i solve this issue? Log error

eleurent commented 3 years ago

Hi @ShivamArora-SA,

You probably used the same seed for the two training runs? Loading a saved model only loads the policy/value neural networks, but the outputs should be different if you select two different seeds (through the --seed argument of experiments.py) for the policy/environment randomness.