This project is being done as a part of Web and Coding Club, IIT Bombay's Seasons of Code 2021.
In this project, our end goal is to try to implement a Deep Reinforcement Learning Algorithm to play the game of 2048 and get the highest possible score.
- Week 1 : Implemented Expectimax Algorithm to create an Agent to play the 2048 achiving a median score of 10K for tree depth of 3.
- Week 2 : Implemented MDP Planning to create an TicTacToe counteragent (given the moves of the oppponents at each state.
- Week 3 : Implemented Tabular q-learning to create a Snake Xenzia AI for an environement of size 4*4 achieving a median length of 12.2
- Week4 : Implemented Deep q-leearning to create an Agent to play the 2048 using a convNet with 4 conv layers & a fully connected layer achieving a median score of 37K & reaching the 2048 tile in more than 90% episodes
