I'm working on search in Pokemon to create engines/solvers and a library for their development. The first phase of this is creating effective MCTS and AlphaBeta solvers in the perfect info setting. I was advised to create a target to compare the output of my library to, so I wrote a C++ program to independently solve 'bull vs bull'. It's fairly tractable and a perfect solution can be calculated in only a few hours.
A better write up is available on my personal site along with the formatted solutions as compressed txt files.
Be warned that I cleaned up multiple mechanics/minor bugs in the process of creating this and so I'm not totally confident the code is correct. I am sure of my methods though and the source code is here. I would appreciate some scrutinizing eyes!
Correct me if I'm wrong but I believe this is the first instance of 'Strong Solving' in competitive Pokemon, even though the setting is extremely specific. Thanks for reading.
Update 1/9/24:
I noticed that I incorrectly had Blizzard's freeze chance set to 30% instead of 10%. With the former the best move in a healthy Tauros mirror (with freeze clause not active) was to Blizzard. Now the best move is to Body Slam. I think the old result might still be valid since 30% is the rate in the Japanese version so I will keep the data up.
A better write up is available on my personal site along with the formatted solutions as compressed txt files.
Be warned that I cleaned up multiple mechanics/minor bugs in the process of creating this and so I'm not totally confident the code is correct. I am sure of my methods though and the source code is here. I would appreciate some scrutinizing eyes!
Correct me if I'm wrong but I believe this is the first instance of 'Strong Solving' in competitive Pokemon, even though the setting is extremely specific. Thanks for reading.
Update 1/9/24:
I noticed that I incorrectly had Blizzard's freeze chance set to 30% instead of 10%. With the former the best move in a healthy Tauros mirror (with freeze clause not active) was to Blizzard. Now the best move is to Body Slam. I think the old result might still be valid since 30% is the rate in the Japanese version so I will keep the data up.
Last edited:
