Closed ihungalexhsu closed 7 years ago
train.py
will have started a tensorboard server, which you can see by navigating to http://localhost:12345/ in a browser. If you've stopped the run, you can start a new tensorboard server with `tensorboard --logdir /tmp/pong --port 12345'
Thanks @tlbtlbtlb
How about using the model playing a game?
( Like env.rendor() in gym )
You can start training again, and it will use the previous checkpoint. Then use a VNC client such as TurboVNC to connect to localhost:5900
.
I cannot connect locally to localhost:5900. I cannot watch it train visually unless it is through tensorboard
Excuse me, I've completed training a model of pong game(using command : python train.py --num-workers 2 --env-id PongDeterministic-v3 --log-dir /tmp/pong ). How can I use the model checkpoint to see how it performed on the game visually?