rl-example

An example of how to use torch-rl.

Black Jack

BlackJack is an example MDP that gives you an idea of how to implement a non-trival MDP.

Some simple scripts you can run:

• visualize_montecarlo_results.lua - See the Q-function and policy that Monte Carlo Learns.
• test_montecarlo.lua - See how Monte Carlo Control does on BlackJack and TestMdp.
• test_tablesarsa.lua - See how Table-Lookup Sarsa Lambda does on BlackJack and TestMdp.
• test_thresholdpolicy.lua - See how a super simple policy does on BlackJack.

This MDP is based on Easy 21 by Johannes.

Visualization

The plot below shows the (greedy) policy that Monte Carlo learns. It shows what the best action (z axis) to take is given a state (xy axis)A value of 1 means the player should "hit", and 2 means that the player should "stick".

The plot below shows the learned value of different states.