Central impl repository for our 8-week Reinforcement Learning project. Our focus is teaching a computer to play Hearts using reinforcement learning.

many index.ts files contain comments explaining the content of the directory.

Agents of all varieties (random, contextless, rule-tracking, contextual, heuristic, and co.) are located in src/agents along with training and evaluation code. The test files demonstrate how to use all the code and directly import all the relevant code. Note the createAgent, gameSummary, and learningMethods modules in particular- they contain the code for leaning methods and feature constructions.

The simulator is located in src/simulator. Reading the tests is probably the quickest way to get a handle on what's going on there if you're curious.

The neural nets are conspicuously located in src/neural-net. Note that the neural net code has since been copied over to another repository and published on npm as the package combobulate.