An attempt at making an AI for Super Smash Bros. Melee using Q-learning
-
First of all install CMake:
Download cmake-3.13.0-rc1.tar.gz from: https://cmake.org/download/
Run the following commands in terminal:
cd \folder\which\contains\the\.tar.gz\filename tar xvzf PACKAGENAME.tar.gz ./configure make sudo make install
-
Install dolphin:
Run the following commands in terminal:
git clone --single-branch -b memorywatcher https://github.com/altf4/dolphin.git cd \path\to\dolphin\folder\just\cloned mkdir build cd build cmake .. make sudo make install
-
Legally acquire an ISO picture of Melee v1.02 NTSC.
-
Clone this repository
-
Run q_learning.py to start the learning!
In order to test the trained agent against a random agent, make a copy of your Q table with a different name, and input that name into the "stored_filename" variable in q_table_benchmarking.py, and start the script.
In order to graph the data generated under the learning, set the parameter "stored_filename" in the call to show_all() to the name of the "stored_filename" variable used in the learning process, and start the script