Q&A of Single Player Learning Track

This page lists some Q&A of the Single Player Learning Track.

---

Will there be a deep learning library available in the machines running the learning track? Such as Tensorflow or DL4J?

Answer: TensorFlow has been installed for Python3.5 on the server machine.

---

On the website, you mention the agent has access to the game state. Does that involve the events that happen or only objects and their positions?

Answer: The game state include the same information as the StateObservation in the planning tracks except the events. The SerializableGameState.java can be found here.

---

During the validation, is there a limit to how much time can be taken per action (like the 40ms limit in the Planning track)? If so, is 'wait' or 'no action' a valid action?

Answer: During the validation, the time limits are the same as during testing, and the same as in the planning tracks. The legal actions are the same as in the planning tracks: ACTION_NIL, ACTION_UP, ACTION_LEFT, ACTION_DOWN, ACTION_RIGHT, ACTION_USE and ACTION_ESCAPE. Please refer to Types.java for Java Client or Types.py for Python Client.

---

For more information, please check:

Single Player Learning Track Specifications

Create an agent

Test an agent