def run_step(step_num):
if step_num == 0:
model = models.RandomModel()
elif step_num == 1:
model = models.PriorModel()
elif step_num == 2:
model = models.SupModel()
elif step_num == 3:
print('Training will take 6-7 minutes')
model = models.EmbedModel()
elif step_num == 4:
print('Training will take 30-40 minutes')
gru = models.GRU()
model = models.NeuralModel(gru)
else:
raise ValueError('Invalid step number')
trainset = Dataset.get('train')
num_train_candidates = Candidate.get_count()
model.fit(trainset, num_train_candidates)
print('Training finished!')
for dsname in Dataset.ds2path.keys():
ds = Dataset.get(dsname)
pred_cids = model.predict(ds)
print(dsname, ds.eval(pred_cids))
def main():
args = docopt(__doc__, options_first=True)
available_models = {
"user": models.UserWasdModel(),
"random": models.RandomModel(),
"knn": models.KnnModel()
}
width = int(args["--width"])
height = int(args["--height"])
if args["--list-models"]:
print("user, random, knn")
elif args["play"]:
log = create_log_file(args["--output"]) if args["--save"] else None
simulate(width, height, models.UserWasdModel(), log, display=True)
elif args["simulate"]:
for _ in range(int(args["NUM_GAMES"])):
log = create_log_file(args["--output"]) if args["--save"] else None
simulate(width, height, available_models[args["--model"]], log,
args["--display"])
for i in range(7):
maze[16][11+i]=1
maze[17][11]=1
game = Maze(maze)
print(maze.size)
# only show the maze
if test == Test.SHOW_MAZE_ONLY:
game.render(Render.MOVES)
game.reset()
# play using random model
if test == Test.RANDOM_MODEL:
game.render(Render.MOVES)
model = models.RandomModel(game)
game.play(model, start_cell=(0, 0))
# train using tabular Q-learning
if test == Test.Q_LEARNING:
game.render(Render.TRAINING)
model = models.QTableModel(game, name="QTableModel")
h, w, _, _ = model.train(discount=0.90, exploration_rate=0.10, learning_rate=0.10, episodes=500,
stop_at_convergence=True)
# train using tabular Q-learning and an eligibility trace (aka TD-lamba)
if test == Test.Q_ELIGIBILITY:
game.render(Render.TRAINING)
model = models.QTableTraceModel(game)
h, w, _, _ = model.train(discount=0.90, exploration_rate=0.10, learning_rate=0.10, episodes=200,
import models
from dataset import Dataset, KnowledgeBase
if __name__ == '__main__':
model = models.RandomModel()
trainset = Dataset.get('train')
model.fit(trainset)
print('【Random Model】')
for dsname in Dataset.ds2path.keys():
ds = Dataset.get(dsname)
pred_cids = model.predict(ds)
print(dsname, ds.eval(pred_cids))
print('--------------------')
model = models.PriorModel()
trainset = Dataset.get('train')
model.fit(trainset)
print('【Prior Probability Model】')
for dsname in Dataset.ds2path.keys():
ds = Dataset.get(dsname)
pred_cids = model.predict(ds)
print(dsname, ds.eval(pred_cids))
print('--------------------')
model = models.SupModel()
trainset = Dataset.get('train')
model.fit(trainset)
print('【Supervised Model】')
for dsname in Dataset.ds2path.keys():
ds = Dataset.get(dsname)