Skip to content

This repository contains Reinforcement Learning course projects...

License

Notifications You must be signed in to change notification settings

Taabannn/intro-rl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

intro-rl

This repository contains Reinforcement Learning course projects...

The goal of HW1 is getting familiar with python and the concepts of Statistical Inference by comparing three different policies.

In HW2, the average received reward by epsilon-greedy, gradient-based and UCB agents with different reward distribution are compared. The impact of tunning hyperparameters like epsilon (in epsilon-greedy algorithm) and learning rate (in UCB) on the average received reward has been observed as well.

Policy Iteration and Value Iteration policies (with respect to their convergence rate) are compared and the impact of threshold and discount factor on their convergence has been observed in HW3.

In HW4, the difference in convergence rate and steady state value of Q-Learning (with constant and decaying learning rate), SARSA, Tree Back up n-step and off-policy Monte Carlo with epsilon greedy behavior policy (in both constant and decaying epsilon) algorithms have been observed.

The main focus of HW5 was on Deep Reinforcement Learning algorithms and for this purpose, DQN algorithm with image observation by using CNN has been implemented and the impact of Transfer Learning in comparison to Ordinary Learning has been cleared.