Make Training Work

[Bonifatius94]

• Put neural net framework in place
  • [x] can solve regression task (e.g. predict sin(x))
  • [x] can solve classification task (e.g. learn MNIST)
• Implement RL algorithms
  • [ ] Deep Q
  • [x] PPO
  • [ ] Monte Carlo Tree Search
• Prepare training data based on gameplay
  • [x] model a reward function
  • [x] model the input features seen by the agent
• Implement Benchmark
  • [x] random play benchmark
  • [ ] benchmark against heuristic agent
  • [ ] benchmark against best trained agent
  • [ ] benchmark with real gameplay data