neural-network-training.js

const discountFactor = 0.85
const trainingSets = []

for (let i = 0; i < numMoves - 1; i++) {
  let currentNetStateNormalized = netConfig.netNormalizedOutput(moves[i].boardVector)[0]
  let reward = moves[i].reward
  let nextNetStateNormalized = netConfig.netNormalizedOutput(moves[i + 1].boardVector)[0]

  let netOutput = normalizeReluOutput(reward + discountFactor * nextNetStateNormalized)

  trainingSets.push({
    boardVector: moves[i].boardVector,
    netOutput: [netOutput]
  })
}

netConfig.net.train(_.map(trainingSets, function (trainingSet) {
  return {
    input: trainingSet.boardVector,
    output: trainingSet.netOutput
  }
}), {
  iterations: 1
})