def run_model(params):
# https://stackoverflow.com/questions/11526975/set-random-seed-programwide-in-python
# https://stackoverflow.com/questions/30517513/global-seed-for-multiple-numpy-imports
random.seed(params.seed)
np.random.seed(params.seed)
# Must be called before Session
# https://stackoverflow.com/questions/38469632/tensorflow-non-repeatable-results/40247201#40247201
tf.set_random_seed(params.seed)
qagent = QAgent(params)
if params.is_train:
qagent.fit()
elif params.eval_mode == 0:
qagent.evaluate_mine()
elif params.eval_mode == 1:
qagent.test_mine()
elif params.eval_mode == 2:
for mines in range(1, 13):
params.mines_min = mines
params.mines_max = mines
print("Mines =", mines)
qagent.test_mine()
tf.reset_default_graph()
qagent = QAgent(params)
def __init__(self):
#display attributes
self.clock = pygame.time.Clock()
pygame.init()
self.font=pygame.font.Font(None,30)
self.size = (1024, 768)
self.size_vec = Vector2(1024, 768)
self.screen = pygame.display.set_mode(self.size)
self.colors = {'WHITE':(255,255,255), 'red': (255,0,0), 'blue': (0,0,255), 'black': (0,0,0)}
#world attr
self.g=980.0#cm/sec^2
#peg-att
self.peg=Vector2(512.0,100.0)
#ball-att
self.ball_length=100.0
#initial state: config
self.ball_theta=m.pi/2#[0,2*pi]
self.ball_omega=0.0
self.ball_alpha=self.g/self.ball_length*m.sin(self.ball_theta)
self.ball_theta_min = 10000
self.ball_theta_max = -10000
self.ball_omega_min = 10000
self.ball_omega_max = -10000
self.ball=Vector2(self.polar_cart())
#self.ball(x,y)
self.player = QAgent(self.get_ranges())
def train(params):
# https://stackoverflow.com/questions/11526975/set-random-seed-programwide-in-python
# https://stackoverflow.com/questions/30517513/global-seed-for-multiple-numpy-imports
random.seed(params.seed)
np.random.seed(params.seed)
# Must be called before Session
# https://stackoverflow.com/questions/38469632/tensorflow-non-repeatable-results/40247201#40247201
tf.set_random_seed(params.seed)
qagent = QAgent(params)
if params.is_train:
qagent.fit()
else:
qagent.test_mine()
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import tensorflow as tf
from agent import QAgent
from configs import object_seaquest_config
from util import get_log_dir
if __name__ == '__main__':
config = object_seaquest_config
log_dir = get_log_dir('log', config['game']+'_'+str(config['double_q'])) # Name of logging directory
agent = QAgent(config=config, log_dir=log_dir)
saver = tf.train.Saver(max_to_keep=None)
saver.restore(agent.session, '%s/episode_%d.ckpt'%("log/log/2017-12-09_23-40-34_SeaquestDeterministic-v4_True",800))
print('Validate....\n==============')
scores = agent.validate_episode(epsilon=0, visualise=True)
from plot_util import init_figure, update_figure
import tensorflow as tf
import numpy as np
from agent import QAgent
from configs import pong_config, breakout_config
if __name__ == '__main__':
config = pong_config
config['state_memory'] = 1 # prevent allocating of a huge chunk of memory
load_episode = 3650
epsilon = 0.05 # The epsilon for the strategy
# Build the graph on CPU to keep gpu for training....
with tf.device('/cpu:0'):
agent = QAgent(config=config, log_dir=None)
# Restore the values....
tf.train.Saver().restore(
agent.session,
'saves/%s/episode_%d.ckpt' % (config['game'], load_episode))
mean = 0
total = 0
episode = 0
while True:
print("\n")
# Initialise the episode
state = agent.reset_to_zero_state()
done = False
total_reward = 0.
import gym
from agent import QAgent
env = gym.make('CartPole-v0')
agent = QAgent(env)
agent.train()
t = agent.run()
print("Time", t)
DISCOUNT_FACTOR = 0.6
SAVE_MODEL_EVERY = 0
if __name__ == '__main__':
#create env
env = gym.make(ENV_NAME)
print(env.unwrapped.spec.id)
#create agent
model = QTable(nostates=env.observation_space.n,
noactions=env.action_space.n,
learning_rate=LEARNING_RATE,
discount_factor=DISCOUNT_FACTOR)
agent = QAgent(actions=env.action_space.n,
expl_max=EXPLORATION_MAX,
expl_min=EXPLORATION_MIN,
expl_decay=EXPLORATION_DECAY,
model=model)
#get and parse user args
args = Parser.parseargs(defaultTrainIterations=10000,
defaultEvalIterations=10)
if args.load:
agent.load(env, args.loadversion)
if args.train != 0:
agent.train(env,
iterations=args.train,
train_s=1,
save_i=SAVE_MODEL_EVERY)
if args.eval != 0:
print("Evaluation results (lower scores are better):")
action='store_true',
help='Use ICM module')
args = parser.parse_args()
env = gym.make(args.gym_env)
if (type(env.action_space) == Discrete):
if args.use_DQN:
a = QAgent(epsilon_start=args.epsilon_start,
epsilon_end=args.epsilon_end,
epsilon_anneal=args.epsilon_anneal,
nb_actions=env.action_space.n,
learning_rate=args.learning_rate,
gamma=args.gamma,
batch_size=args.batch_size,
replay_memory_size=args.replay_memory_size,
hidden_size=args.hidden_size,
model_input_size=env.observation_space.shape[0],
use_PER=args.use_PER,
use_ICM=args.use_ICM)
trainQ(a, env, args.MAX_NUMBER_OF_STEPS, args.EPISODES_TO_TRAIN,
args.START_RENDERING, args.update_frequency)
else:
if not args.use_ICM:
a = ActorCriticAgent(
continuous=False,
nb_actions=env.action_space.n,
learning_rate=args.learning_rate,
gamma=args.gamma,
hidden_size=args.hidden_size,