/unicorn-reinforcement-learning

Deep reinforcement learning for the toggl's Unicorn Startup Simulator

Primary LanguageJavaScriptMIT LicenseMIT

unicorn-reinforcement-learning

Deep reinforcement learning for the toggl's Unicorn Startup Simulator

https://toggl.com/startup-simulator/

This project uses the excellent ConvNetJS javascript library for training the neural network entirely in the browser.

The trained network has 80% plus win rate.

Instructions

To run the trained network:

The agent will start to answer the questions every 3 seconds.

Training

To train the network you can follow the same steps but instead use the agent-train.js file. The training can take a lot of time depending on your hardware. After the training finishes a 1000 game evaluation will follow. Finally the trained network will be copied in your clipboard. You can paste it in a file for further use.