Hello World, today (August 25, 2020). I trained an agent to play a navigate successfully a game called frozen lake

The game can be found at https://gym.openai.com/

DISCOVERIES:

learing rate:

setting learning rate at 0.1 the qtable change by very little and the accumulated reward was also low

setting learning rate at 0.7 increasing accumulated reward significantly , but the qtable changed rapidly

setting learning rate at 0.5 the agent achieve a good amount of accumulated reward and the q table changed rapidly

number of episodes:

increasing the number total number of episode generally increased the accumulated reward and change in the qtable was significant.

reinforcement learning: q-learning

using environments in the gym package https://gym.openai.com/

implementing epsilon greedy algorithm with exponential decay

I employ you to implement the https://github.com/ibkvictor/frozen-lake-agent/blob/master/frozen_lake.ipynb and learn something cool.

please encourage me by forking this repo and giving it a star.

Your friend

Victor.