Skip to content

Latest commit

 

History

History
34 lines (27 loc) · 910 Bytes

README.md

File metadata and controls

34 lines (27 loc) · 910 Bytes

Reinforcement_Project

Installation

  • !pip install gym
  • !pip install pyglet
  • !pip3 install box2d-py
  • !pip3 install gym[Box_2D]

Contributors

  • Abdelkerim Boukadida
  • Achref Souibgui

Project : Car-Pole

In this task we have to balance a rod on top of a cart. Number of action spaces is 2. Action space is discrete here.

  • 0 - move cart to the left
  • 1 - move cart to the right

we used :

  • DQN for our q learning algorithm
  • MSE loss function
  • Adam optimizer

About

A pole is attached by an un-actuated joint to a cart, which moves along a frictionless track. The system is controlled by applying a force of 0 or 1 to the cart. The pendulum starts upright, and the goal is to prevent it from falling over.

Hyperparameters:

  • Gamma= 0.95
  • LEARNING_RATE=LR=0.001
  • Epsilon=1
  • Epsilon_decay=0.995
  • Epsilon_min=0.01
  • BATCH_SIZE=64