Skip to content

Instantly share code, notes, and snippets.

@AlexandreSi
Last active October 12, 2021 21:29
Show Gist options
  • Save AlexandreSi/4b117e51f0f9c0cf2ef430f6e85787ed to your computer and use it in GitHub Desktop.
Save AlexandreSi/4b117e51f0f9c0cf2ef430f6e85787ed to your computer and use it in GitHub Desktop.
Parameters definition in index.js
// type of neural network to train
const NETWORK_TYPE = 'CNN';
// number of games to play
const LEARN_TIMES = 10000;
// learningRate is progressively decreased with the number of games until
// the final value LR_INIT/LR_FINAL_FRACTION
const LR_INIT = 0.0001;
const LR_FINAL_FRACTION = 10;
// epsilon is the ratio between exploration and exploitation
// it can evolve along the games played
const EPSILON_INIT = 0.1;
const EPSILON_FINAL = 0.4;
// gamma is the fraction attributed to the maximum Q Value of the next state
const GAMMA = 0.3;
// reward awarded to the final play that led to victory
const REWARD = 100;
// discount applied to the reward awarded to the previous plays that led to victory
const DISCOUNT = 0.8;
// reward (or - reward) awarded to prevent the bot to lose
const SIDE_REWARD = 75;
// network configuration
const NETWORK_CONFIG = {
CNN: {
filters_number: 32,
padding: 2,
stride: 1,
size: 4,
layers : [60, 30]
},
NN: {
layers: [100]
}
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment