x = tf.reshape(x, [-1, 7 * 7 * 64], name='flatten')
x = tf.layers.dense(
x, units=1, activation=None, name='fc1'
) # sigmoid is applied implicitly in the computation of the loss
return x
if __name__ == '__main__':
_BATCH_SIZE = 128 # Size of the minibatches
_CODE_SIZE = 10 # Size of the latent code
_MAX_IT = 20000 # Number of iterations
_LOG_BASE_DIR = 'log'
_LOG_DIR_SUFFIX = 'mnist_gan'
_LOG_DIR = get_log_dir(_LOG_BASE_DIR, _LOG_DIR_SUFFIX)
_CHECKPOINT_FILE = '%s/model.ckpt' % _LOG_DIR
# Load mnist data and create a function to get the batches
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
mnist_batch = lambda x: mnist.train.next_batch(x)[0].reshape(x, 28, 28, 1)
with tf.Session(config=tf.ConfigProto(log_device_placement=True)) as sess:
# Create the network and initialise the parameters
gan = DCGAN([28, 28, 1], _BATCH_SIZE, _CODE_SIZE)
sess.run(tf.global_variables_initializer())
# Create file writers and summary
merged = tf.summary.merge_all()
saver = tf.train.Saver()
train_writer = tf.summary.FileWriter(_LOG_DIR, sess.graph)
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 __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 pong_config, breakout_config
from util import get_log_dir
if __name__ == '__main__':
config = breakout_config
log_dir = get_log_dir('log', config['game']+'_'+str(config['double_q']))
agent = QAgent(config=config, log_dir=log_dir)
saver = tf.train.Saver()
for episode in range(config['episodes']):
print('\n\nepisode: %d, step: %d, eps: %.4f\n\n---------------------' % (episode, agent.steps, agent.epsilon))
# Store the rewards...
agent._update_training_reward(agent.train_episode())
if episode % config['episodes_validate']==0:
print('Validate....\n==============')
scores = [agent.validate_episode(epsilon=0.05) for i in range(config['episodes_validate_runs'])]
agent._update_validation_reward(np.mean(scores))
print(scores)
# Store every validation interval
if episode % config['episodes_save_interval']==0:
saver.save(agent.session,'%s/episode_%d.ckpt'%(log_dir,episode))