• The README is really clear and appreciated
• Your way of doing alpha beta pruning without alpha and beta variables was really smart and simple, good job!
• The use of a gaussian distribution to reinforce the idea of exploration vs exploitation was a really creative solution
• A better commenting coverage for the GA file would be helpful
• For the GA strategy, having an offspring of 7 with population 10 is an interesting choice. Have you tried different values or was this the best ones?
• Your results for the GA agent are really good, especially the win rate against the optimal agent

General Comments

I really appreciate your readme file, it's very clear. Also, your heuristic is very good, and in many case converges to the global optimum(5, 10, 20).

I also agree wiht you to find a trade-off between exact solution and computation complexity.

Code suggestion

• I didin't understand well your code, some comments can help.
• Why you used a prioritized set? And why you inserted it into a priority queue? Didn't was enough to put into the priority queue?

Lab2 Review Luca Villani for Lorenzo Bellino

You have implemented a very good solution and you had some very good intuition. The code is well written and working

Some Comments

Some possible improvement and advice that i can give you:

• You can try to remove duplicates from the population, it would decrease your computational time and lead to a more efficient way to manage the steady state. (Even tough, the recombination after 7 steady is a very good idea!)
• About the mutation rate, which in your code is fixed, you can try to implement a dynamic system to manage it. It could be a good idea to balance exploration and exploitation.
• Before mutating some of the offsprings you can try to check them, and mutating only the worst ones

The code is unfortunately poorly commented but the code itself is readable

General impression

I think you have done a good job of finding a solution with the tradeoffs you presented in the readme. Your readme is highly appreciated, and I like that it is thoroughly written.

Small suggestion

• Some parts of your code could become easier to understand by dividing it into smaller steps with good variable names. For example, I would suggest dividing the line presented below into sublines for a better understaning, perhaps declaring "int(100*len(element)/N" as a varible.

 options.put(PrioritizedSet(int(100*len(element)/N),element))

• Good use of docstrings and I think it the code could be more easily understandable with some more comments in your search-function.

• Good idea to implement the weight as percentage of new coverage in each set.

Hey Lorenzo!
First of all, I want to tell you that I really appreciate the completeness and readability of Readme: it helped me a lot to understand your choices and your implementation.

MAJOR

• I agree with you about the fact that a greedy strategy could be the best tradeoff between amount of computation time and a reasonable (maybe not optimal) solution, especially with larger values of N. But when you facing the problem for N<=20, other implementations (like A* search) allow us to find an optimal solution without a huge amount of computation.

###MINOR

• You can try to speed up your search, in general, by removing duplicates of the elements of the initial list of lists, and remove the list who are empty.

General notes

• Thoroughly written README and I appreciate the links to the files under each heading.
• In general, I am inspired by your solutions and think you managed to create good solutions.

Task 3.1

• I see that you have created agents of different levels but I was wondering if you tried to create an agent with fixed rules that performed better/as well as the pure_random and/or the optimal strategy?

Task 3.2

• Well done. I am inspired by your solution and think you managed to create a good solution.
• I like that you are playing your GA against different opponents of different levels.
• Did you play around with the variables like population_size, generations, etc?
• Although I tried to run "nim_GA.py" but could not seem to make it work since in line 25 you are calling opponents.cook_data(state) but opponents do not have a function cook_data.

Task 3.3

• Well done and good that you have improved the basic minmax with the memory, pruning and max_depth. Did you try different max_depth for different Nim_sizes?
• I like that you have good explanations of your functions and I think it would have been even better with a recommendation for the max_depth for different nim_sizes.

Task 3.4

• You did not finish the RL agent.