qnet Namespace Reference

Classes
class	QValueNetwork
	— Q-value networks More...

Functions
def	disturb (u, i)
def	onehot (ix, n=NX)
def	rendertrial (maxiter=100)

Variables
float	DECAY_RATE = 0.99
	env = DPendulum()
	— Environment
	feed_dict
list	h_rwd = []
	— History of search
float	LEARNING_RATE = 0.1
int	NEPISODES = 500
	— Hyper paramaters
int	NSTEPS = 50
	NU = env.nu
	NX = env.nx
	optim
	Q2 = sess.run(qvalue.qvalue, feed_dict={qvalue.x: onehot(x2)})
	Qref = sess.run(qvalue.qvalue, feed_dict={qvalue.x: onehot(x)})
	qvalue = QValueNetwork()
	RANDOM_SEED = int((time.time() % 10) * 1000)
	— Random seed
	reward
float	rsum = 0.0
	sess = tf.InteractiveSession()
	u = sess.run(qvalue.u, feed_dict={qvalue.x: onehot(x)})[0]
	x = env.reset()
	— Training
	x2

Detailed Description

Example of Q-table learning with a simple discretized 1-pendulum environment using a
linear Q network.

Function Documentation

onehot()

def qnet.onehot	(		ix,
			n = NX
	)

Return a vector which is 0 everywhere except index <i> set to 1.

Definition at line 58 of file qnet.py.