PyReinforce

PyReinforce is a collection of algorithms that can be used to solve a variety of different reinforcement learning problems.

Breakout	Pong	Space Invaders

Cart Pole	Lunar Lander	Pendulum

Basics

This is how OpenAI Gym describes the reinforcement learning process:

There are two basic concepts in reinforcement learning: the environment (namely, the outside world) and the agent (namely, >the algorithm you are writing). The agent sends actions to the environment, and the environment replies with observations and >rewards (that is, a score).

PyReinforce is fully compatible with OpenAI Gym. In fact, it will work with any environment as long as it implements these methods:

• reset()
• step(action)

Your part is to implement a Brain: a neural network that agents use to decide which actions to pick for given states. Agents get better over time by performing training steps on their brains. See examples for more details.

Installation

Install PyReinforce from PyPI:

pip install PyReinforce

or from source:

git clone https://github.com/aserhiychuk/pyreinforce.git
cd pyreinforce
pip install -e .

Examples

• Monte Carlo
• Temporal Difference
• Policy Gradient
• Deep Deterministic Policy Gradient
• Deep Q-Learning

In order to run the examples you need to install dependencies:

pip install -r examples/requirements.txt