/sarsa

SARSA - Reinforcement Learning

Primary LanguageJupyter Notebook

sarsa

SARSA

Sarsa implemented from scratch in Python. Inspired by real-life neuroscience experiments run on rats. This experiment simulates a rat put on a table with a piece of cheese. Falling off the table is a negative outcome and finding the cheese is positive.

Implementation details:

  • continuous state space
  • function approximation on state
  • eligibility trace

The report can be found at hobbs.im and presents implementation details, effects of various model parameters, limitations of the algorithm, and possible extensions to the project.