Retraining a Model

Rough notes

In the (rl_deepracer_coach_robomaker.py)[https://github.com/crr0004/deepracer/blob/master/rl_coach/rl_deepracer_coach_robomaker.py#L115] file on line 115, uncomment both lines. This will set it to look for checkpoints in that bucket and the folder. In the folder that minio is using to serve your data, in the bucket, create a folder called rl-deepracer-pretrained, and in that folder, create a folder model, so it the whole path looks like bucket/rl-deepracer-pretrained/model.

Now copy the *.chk* and checkpoint files from the currently trained model folder into the pretrained model folder. When launching the sagemaker container you should see a log entry on the lines of Using pretrained checkpoint: ....

I will update this with proper steps and screenshots soon, open an issue if you're not sure what to do in the meantime.