MCarlomagno

var dispatchAction = function (action) {
  var pressed = {},
    self = this,
    f1 = mk.game.fighters[0],
    f2 = mk.game.fighters[1];

  // where pressed is a map of type <keyboard_code, boolean>
  pressed[keyCode] = true;
  var move = self._getMove(pressed, mk.controllers.keys.p1, 0);
  self._moveFighter(f1, move);

MCarlomagno

var playGame = async function () {
  var subzeroIndex = 0;
  var kanoIndex = 1;
  
  startGame();
  var state = getState(subzeroIndex);

  // while both fighters keep alive.
  while (state.myLife !== 0 && state.opponentLife !== 0) {
    

MCarlomagno

    env = gym.make('CartPole-v1')
    state_size = env.observation_space.shape[0]
    action_size = env.action_space.n
    agent = DQNAgent(state_size, action_size)
    # agent.load("./save/cartpole-ddqn.h5")
    done = False
    batch_size = 32

    for e in range(EPISODES):
        state = env.reset()

MCarlomagno

var subzeroMemory = [];
var kanoMemory = [];
var subzeroIndex = 0;
var kanoIndex = 1;
var gamma = 0.95;    // discount rate
var epsilon = 1.0;   // exploration rate
var epsilonMin = 0.01; 
var epsilonDecay = 0.99;
var learningRate = 0.001;

MCarlomagno

var train = async function () {
  var done = false;
  var state = {};
  var time_step = 0;
  var batchSize = 32;
  var subzeroAction;
  var kanoAction;
  var subzeroLife = 100;
  var kanoLife = 100;
  var done = false;

MCarlomagno

var buildModel = function () {
  var model = tf.sequential();
  model.add(tf.layers.dense({ units: 24, inputShape: [6], activation: 'relu' })); // input shape === state shape
  model.add(tf.layers.dense({ units: 24, activation: 'relu' }));
  model.add(tf.layers.dense({ units: 9, activation: 'linear' }));  // units === action_size
  model.compile({ loss: tf.losses.huberLoss, optimizer: tf.train.adam(learningRate) });
  return model
}

MCarlomagno

var updateTargetModel = function () { 
  // copy weights from model to target_model
  for (let i = 0; i < model.layers.length; i++) {
    targetModel.layers[i].setWeights(model.layers[i].getWeights());
  }
}

MCarlomagno

var predictSubzeroAction = function (state) {
  if (Math.random() <= epsilon) {
    return randomAction(subzeroActions);
  }
  var inputState = stateToTensor(state, subzeroIndex);
  output = model.predict(inputState).dataSync();
  return output.indexOf(Math.max(...output));
}

MCarlomagno

var memorize = function (state, action, reward, nextState, done, memory) {
  memory.push({state, action, reward, nextState, done});
} 

MCarlomagno

var underscore = require('underscore');
var replay = async function (batchSize, index) {
  var minibatch = [];

  if (index === subzeroIndex) { 
    minibatch = underscore.sample(subzeroMemory, batchSize);
  } else {
    minibatch = underscore.sample(kanoMemory, batchSize);
  }