Resolve "deterministic actions when calling play ()"

Summary

Added deterministic actions for agent evaluation with play() function.

Explain your context

When training, the agent picks non deterministic actions, to explore the environment and gain experience. When the learning process is finished and the agent is applied to the system, only forword passes are performed with the policy, potentially in the real system. Then the agent should always pick the action with the highest probability to ensure robust application and no random and not ideal actions. That is why the play() function should call the model.predict() with deterministic actions activated.

Please list breaking changes, new features or bugfixes

• Bugfixes: setting deterministic flag to True when calling model.predict()

Closes #291 (closed)