def main():
log.info('Loading %s...', Game.__name__)
g = Game(8)
log.info('Loading %s...', nn.__name__)
nnet = nn(g)
if args.load_model:
log.info('Loading checkpoint "%s/%s"...', args.load_folder_file)
nnet.load_checkpoint(args.load_folder_file[0], args.load_folder_file[1])
else:
log.warning('Not loading a checkpoint!')
log.info('Loading the Coach...')
c = Coach(g, nnet, args)
if args.load_model:
log.info("Loading 'trainExamples' from file...")
c.loadTrainExamples()
run = wandb.init(project=args.wandb_project,config=args,reinit=True)
log.info('Starting the learning process 🎉')
c.learn()
run.finish()
from utils import *
args = dotdict({
'numIters': 1000,
'numEps': 100,
'tempThreshold': 15,
'updateThreshold': 0.6,
'maxlenOfQueue': 200000,
'numMCTSSims': 25,
'arenaCompare': 40,
'cpuct': 1,
'checkpoint': './temp/',
'load_model': False,
'load_folder_file': ('/dev/models/8x100x50', 'best.pth.tar'),
'numItersForTrainExamplesHistory': 20,
})
if __name__ == "__main__":
g = Game(6)
nnet = nn(g)
if args.load_model:
nnet.load_checkpoint(args.load_folder_file[0],
args.load_folder_file[1])
c = Coach(g, nnet, args)
if args.load_model:
print("Load trainExamples from file")
c.loadTrainExamples()
c.learn()
def play_previous_generations(player,
folder,
num_matches,
game,
tempThreshold,
savefolder=False):
#
wandb.init(project="8x8 against previous iterations")
wins = []
losses = []
draws = []
names = []
i = 0
generation = 0
while i <= 70:
print(folder + "checkpoint_" + str(i))
if os.path.isfile(folder + "checkpoint_" + str(i)):
print("loading generation " + str(generation))
network = nn(game)
network.load_checkpoint(folder=folder,
filename="checkpoint_" + str(i))
args = dotdict({
'numMCTSSims': player.args.numMCTSSims,
'cpuct': player.args.cpuct
})
neuralplayer = NeuralNetworkPlayer(game, network, args)
neuralplayer.name = "Neural Network Generation " + str(
generation)
names.append(neuralplayer.name)
if savefolder != False:
save = savefolder + "/" + neuralplayer.name + "/"
else:
save = False
arena = Arena(player,
neuralplayer,
game,
tempThreshold=tempThreshold)
winsG, lossesG, drawsG = arena.playGames(num_matches,
save=save)
wins.append(winsG)
losses.append(lossesG)
draws.append(drawsG)
wandb.log({
'Wins':
winsG,
'Losses':
lossesG,
'Draws':
drawsG,
'Winrate':
winsG / (winsG + drawsG + lossesG),
'Winrate including draws':
(winsG + 0.5 * drawsG) / (winsG + drawsG + lossesG)
})
generation = generation + 1
i = i + 1
df = pd.DataFrame(wins, columns=player.name, index=names)
df2 = pd.DataFrame(wins, columns=player.name, index=names)
if savefolder != False:
df.to_csv(r"" + savefolder + "/wins.csv")
df2.to_csv(r"" + savefolder + "/draws.csv")
'numEps': 2, #orginal value: 100
'tempThreshold': 15,
'updateThreshold': 0.6,
'maxlenOfQueue': 200000,
'numMCTSSims': 25,
'arenaCompare': 40,
'cpuct': 1,
'checkpoint': './temp/',
'load_model': False,
'load_folder_file': ('/dev/models/8x100x50', 'best.pth.tar'),
'numItersForTrainExamplesHistory': 20,
})
if __name__ == "__main__":
g = Game(6) #returns the game object (constructor)
nnet = nn(g) #NNet class returns NNetWrapper for the game object (g)
print(
'----------------------********************-----------------------*********************-----------------'
)
if args.load_model:
nnet.load_checkpoint(args.load_folder_file[0],
args.load_folder_file[1])
print('main.py==> ', 'args: ', args)
c = Coach(
g, nnet, args
) #returns the Coach object with params(game_object, NeuralNet, argument values)
if args.load_model:
print("Load trainExamples from file")
c.loadTrainExamples()
c.learn()