on
Reinforcement Learning Algorithms
Spinning Up is a great resource by Open AI for Deep Reinforcement Learning. One of the suggestions they have is to code some of the core algorithms from scratch.
Spinning Up is available here: https://spinningup.openai.com/en/latest/index.html
This blog post “A (Long) Peek into Reinforcement Learning” is another useful resource that explains many of the algorithms here: https://lilianweng.github.io/lil-log/2018/02/19/a-long-peek-into-reinforcement-learning.html#q-learning-off-policy-td-control
Since many resources already exist, I plan to mostly just use this as a place to link to the code that I’ve written, but will also mention any useful points that I come across.
Tabular Q
Double Q
SARSA
DQN
Rainbow
VPG
Possibly more efficient to represent policy e.g. value hard to calculate, but know if ball going here then go there don’t need max (pushes majorly in that direction)