Analyzing issues with Player Zero (3/3)

Explain the context of the issue, what is being addressed in detail.

The goal

[x] Verify training works with a simplified game
- [x] Value training works
[v] Verify obstacles properly represented in the game output
[x] Check if we account for exploit multiplier when converting Tree Score to policy
[x] Training with higher params (N simulations, Max node depth for serialization)
[x] replace convergence detector to monitor accuracy instead of loss
[x] better plots for model training visuals
[x] null loss after first generation
- caused by null policies in the MCTS
- fixed by clamping the values with epsilon to avoid division by 0
[x] meteor can cause a state where both players are killed (Tree leaf)
[x] Over-clamping values during node serialization

Time Estimate: 0 hours 0 minutes Time spent: 8 hours 47 minutes

...