We tried to deploy the baseline solution by applying the two files of the trainer and policy scripts to train our models, but with the baseline solution trainer, the panda window runs infinitely without writing any models in the targetted directory. How can this problem be fixed? Thanx.
Posted by: Team115 @ Dec. 9, 2019, 1:06 a.m.Thanks for your message. Did you ensure that the 'stop' condition was met? In the gist, its set to:
stop={'episodes_total': 10000},
But this can be adjusted as desired. The training should stop once this condition is met and the model saved. You can also cancel a run prematurely, restore from checkpoint and set the stop condition to a low value, to ensure almost immediate saving of the model, for example. Also, you may want to train in headless mode, which should be faster, using --headless argument.
Posted by: HuaweiUK @ Dec. 9, 2019, 4:23 p.m.