Lab3 peer review

[LeonardoTredese]

Hardcoded player

• You do not need function closures to create the benchmarking strategies, you could have used directly the name of the defined functions.

Evolved strategies

• You could have used an additional parameter p3 for the 2 params evolutive strategies to make the two inner ifs independent.
• cooked["completion"] = sum(o for o in state.rows) / state.total_elements should always be 1 because total_elements is defined as:

  @property
  def total_elements(self) -> int:
      return sum(self._rows)

  so using the data["completion"]==1 is useless.

• I apreciate that you tried a population method as well, but make_strategy_evol is not defined so it crashes, nice that you did a tweak of the parameters, it would have been nice if you added some kind of simple crossover.
• In tournament selection is just a random the choice of an individual, you should look also at their fitness when selecting them

MinMax

• 👌
• min max with ab pruning does not beat optimal strategy like nomal minmax (which already does some pruning if(val==-1 and player==0):)

RL

• 👌
• Nice that you itialize self._q just in time
• Nice that the agent is trained to play first and last
• Nice that you give a negative reward to the opponent's mistakes