Cart Pole experiment
This experiment uses Tensorflow.js in order to solve the cartpole problem using Policy Gradient Reinforcement learning.
Algorithms
You can choose between several algorithms :
- REINFORCE
- REINFORCE with baseline
- Actor Critic (A2C)
Technologies
- Tensorflow.js is used for the neural network
- The Experiment is rendered using Phaser 3 game engine
- Tfjs-vis is used for learning graph visualization
![](http://www.noxis-studio.com/blog/wp-content/uploads/2019/03/cartpole_trained.gif)
The Settings page allows to:
- Configure the Hypermarameters of the Policy and Value models
- Store the trained models in the Browser
- Retrieve and use stored models
![](http://www.noxis-studio.com/blog/wp-content/uploads/2019/03/cartpole_settings-1.png)