def TrainEpoch(self, e):
games = [Game.GetRandomGame(e) for i in range(self.batch)]
OIndexes = [i for i in range(self.batch) if games[i].GetCurrentPlayer() == Game.O]
XIndexes = [i for i in range(self.batch) if games[i].GetCurrentPlayer() == Game.X]
xInput = np.stack([g.AsVector() for g in games], axis=0)
yTrue = self.PlayGames(games)
trainedModel : Model = self.modelO.GetModel()
trainedModel.train_on_batch(x=xInput[OIndexes], y=yTrue[OIndexes])
trainingLossO = trainedModel.evaluate(xInput[OIndexes], yTrue[OIndexes])
print("modelO: training_loss: %f" % trainingLossO)
trainedModel : Model = self.modelX.GetModel()
trainedModel.train_on_batch(x=xInput[XIndexes], y=yTrue[XIndexes])
trainingLossX = trainedModel.evaluate(xInput[XIndexes], yTrue[XIndexes])
print("modelX: training_loss: %f" % trainingLossX)
print("total loss: %f" % (trainingLossO + trainingLossX))
if e % 3 == 0:
print("Example of O: ")
exampleGame = games[OIndexes[0]]
print(exampleGame)
print(np.resize(self.GetPrediction(self.modelO, exampleGame), (Game.ROW_COUNT, Game.COL_COUNT)))
print(np.resize(yTrue[OIndexes[0]], (Game.ROW_COUNT, Game.COL_COUNT)))
print("Example of X: ")
exampleGame = games[XIndexes[0]]
print(exampleGame)
print(np.resize(self.GetPrediction(self.modelX, exampleGame), (Game.ROW_COUNT, Game.COL_COUNT)))
print(np.resize(yTrue[XIndexes[0]], (Game.ROW_COUNT, Game.COL_COUNT)))
if e % 20 == 0:
print("Game play: ")
def GameCallback(g):
print("----------------")
print(g)
model = { Game.O : self.modelO, Game.X : self.modelX }[g.GetCurrentPlayer()]
print(np.reshape(self.GetPrediction(model, g), (Game.ROW_COUNT, Game.COL_COUNT)))
self.PlayGame(Game(), GameCallback)
return trainingLossO, trainingLossX
def RandomGames():
game = Game.GetRandomGame()
assert (game.GetWinner() == Game.EMPTY)
return game
