-
Notifications
You must be signed in to change notification settings - Fork 0
Meeting 27 September 2017
Decidetto edited this page Sep 27, 2017
·
6 revisions
- Joshua Scheidt
- Marciano Geijselaers
- Timo Raff
- Max Meijers
- Simon Craenen
- What track are we going to choose?
Peer review session went okay
GitHub set up. Contains the following:
- Task list
- Wiki pages for meetings minutes
Discussion of deciding on tracks
Learning:
- work with json
- You get some time to train model on levels 1 and 2
- Play levels 3 and 4 several times for validation
- Q-learning
- Might want to use MCTS -- but not necessarily
Planning:
- MCTS is the strongest
- Can be improved
Ideas:
- Give high rewards to things we don't know yet: to encourage exploration
Conclusion:
-
Go for learning track
-
If we can't get a proper agent running, just make a random one!
-
If we can't make an evaluation function, we just make a random one!
-
Let's try to get something running this period, before we make a full decision
- But for now, the Learning Track it is!