Make number of states dynamic

[aileen-reichelt]

In the current coli-agent/callbacks.py (PR #1), there's a TODO in the agent's setup() function to make the number of states, on which the size of the q-table depends, dynamic. One way to do this could be to pass a generic "example" game state dict to state_to_features() and see how long the returned np.array is. This dict could perhaps be stored outside of callbacks.py (don't know where exactly) so that in can also be accessed in tests. Does that seem like a good solution? Or did you already have something in mind @ivo-1? The only potential problem I see with this approach is that, depending on how complicated state_to_features() ends up being, it could take extra time.

[ivo-1]

Maybe we can just pass the initial game state to see how long the returned array is. If state_to_features() takes that long that it's not viable, it's taking too long 😆

[aileen-reichelt]

We currently have a file of which the length == number of states