def reset(self):
if self._filename is None:
self.delay_queue = DelayQueue(0.01)
else:
self.delay_queue = NoloopDelayQueue(self._filename, 0.01)
self.random_video()
self.state = np.zeros((S_INFO, S_LEN))
return self.state
def __init__(self, random_seed=RANDOM_SEED, filename=None):
np.random.seed(random_seed)
self._filename = filename
if self._filename is None:
self.delay_queue = DelayQueue(0.01)
else:
self.delay_queue = NoloopDelayQueue(self._filename, 0.01)
self.simtime = 0.0
self._video = None
self._videocount = 0.0
self._videoindex = 0.0
self._video_vmaf = []
self._video_len = []
self.random_video()
def __init__(self, random_seed=0, filename=None):
#os.system('rm -rf results')
os.system('mkdir results')
np.random.seed(int(time.time()))
_id = np.random.randint(1000)
self._filename = filename
if self._filename is None:
self.delay_queue = DelayQueue(0.01)
else:
self.delay_queue = NoloopDelayQueue(self._filename, 0.01)
self.simtime = 0.0
self._video = None
self._videocount = 0.0
self._videoindex = 0.0
self._video_vmaf = []
self._video_len = []
self.random_video()
self.last_rtt = -1
self.last_vmaf = -1
self.time_stamp = 0.0
self.state = np.zeros((S_INFO, S_LEN))
self.x = tf.placeholder(
shape=[None, INPUT_SEQ, INPUT_H, INPUT_W, INPUT_D],
dtype=tf.float32)
self.y_ = tf.placeholder(shape=[None, OUTPUT_DIM], dtype=tf.float32)
self.core_net = self.vqn_model(self.x)
self.core_net_loss = tflearn.objectives.mean_square(
self.core_net, self.y_)
gpu_options = tf.GPUOptions(allow_growth=True)
self.sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
self.sess.run(tf.global_variables_initializer())
self.saver = tf.train.Saver()
self.saver.restore(self.sess, "model/nn_model_ep_best.ckpt")
self.x_buff = np.zeros([INPUT_SEQ, INPUT_H, INPUT_W, INPUT_D])
self.action_space = spaces.Discrete(A_DIM)
self.observation_space = spaces.Box(0,
5.0, [S_INFO, S_LEN],
dtype=np.float32)
self.last_vmaf_fake = -1
_strtime = time.strftime('%b-%d-%H:%M:%S-%Y', time.localtime())
self.log_file = open(LOG_FILE + '_' + str(_strtime), 'wb')
def __init__(self, random_seed=RANDOM_SEED, filename=None):
np.random.seed(random_seed)
self._filename = filename
if self._filename is None:
self.delay_queue = DelayQueue(0.01)
else:
self.delay_queue = NoloopDelayQueue(self._filename, 0.01)
self.simtime = 0.0
self._video = None
self._videocount = 0.0
self._videoindex = 0.0
self._video_vmaf = []
self._video_len = []
self.random_video()
self.x = tf.placeholder(
shape=[None, INPUT_SEQ, INPUT_H, INPUT_W, INPUT_D],
dtype=tf.float32)
self.y_ = tf.placeholder(shape=[None, OUTPUT_DIM], dtype=tf.float32)
self.core_net = self.vqn_model(self.x)
self.core_net_loss = tflearn.objectives.mean_square(
self.core_net, self.y_)
# + lossL2
# self.core_train_op = tf.train.AdamOptimizer(
# learning_rate=LR_RATE).minimize(self.core_net_loss)
# self.core_net_acc = tf.reduce_mean(
# tf.abs(core_net - y_) / (tf.abs(core_net) + tf.abs(y_) / 2))
# core_net_mape = tf.subtract(1.0, tf.reduce_mean(
# tf.abs(core_net - y_) / tf.abs(y_)))
#train_len = X.shape[0]
#g2 = tf.Graph()
#gpu_options = tf.GPUOptions(allow_growth=True)
self.sess = tf.Session()
self.sess.run(tf.global_variables_initializer())
self.saver = tf.train.Saver()
self.saver.restore(self.sess, "model/nn_model_ep_350.ckpt")
self.x_buff = np.zeros([INPUT_SEQ, INPUT_H, INPUT_W, INPUT_D])