user@devops:~$ cat README.md
Reinforcement Learning
# Description
Complete Reinforcement Learning project with three fundamental approaches. Tabular Q-Learning on FrozenLake-v1 (4x4 grid, 16 states, 4 actions, slippery ice physics, 5,000 episodes, 16x4 Q-table, 15% evaluation success). Deep Q-Network on CartPole-v1 (continuous 4D state space, fully-connected 64-64 neural network, 300 episodes, Experience Replay buffer 50K, Target Network update every 20 episodes, average reward 130). Includes learning curves, optimal grid policy visualization, Q-value heatmap, and Q-Learning vs DQN comparison boxplot.
# Key features
$ Tabular Q-Learning with 16x4 Q-table for FrozenLake-v1
$ Deep Q-Network with 64-64-2 network for continuous 4D CartPole-v1
$ Experience Replay with 50,000 experience buffer
$ Target Network for DQN training stability
$ Epsilon-greedy with progressive decay (1.0 -> 0.01)
$ Learned policy evaluation (50 episodes, avg reward ~130)
$ 5 visualizations: learning curves, heatmaps, policy, boxplot
# Gallery
# Technologies used
