Import model back

[Jack17432]

By Nick & Bowen

PLEASE READ

Change:

• In train.yaml, if actor_path and critic_path are specified, we can continue training with previous trained model.
• Also if they are specified, the exploration steps is 0. Because no point to explore when we have the trained model.
• Added a bash script that restart the training every $1 time ($1 is the input param). After each restart, it will get the latest stored/modified model and then use it to continue training it.

Note:

• To restart the training every $1 time, you HAVE TO run the training via. retrain.sh.
• If either actor_path and critic_path is not specified, then it will run the training like before.