epi = 100
step_limit = 500
avg_steps = []
avg_reward = []
env = gym.make('CartPole-v0')
with tf.Session() as sess:
sess.run(init)
for i_episode in range(epi):
# obs = env.reset()
obs = model.reset()
for step in range(1, model.Nsim + 1):
action_val = action.eval(feed_dict={X: obs.reshape(1, num_inputs)})
action_picked = action_list[action_val[0][0]] #
model.u[step, 0] = model.u[step - 1, 0] + action_picked #
model.x[step, :] = model.next_state(model.x[step - 1, :],
model.u[step, :]) #
obs = model.x[step, :] #
reward = reward_calc(model.x[step, 1], 324.5) #
# obs, reward, done, _ = env.step(action_val[0][0])
trainer = tf.train.GradientDescentOptimizer(learning_rate=rl.learning_rate)
updateModel = trainer.minimize(loss)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
rList = []
with tf.Session() as sess:
sess.run(init)
saver.restore(sess, "/tmp/model.ckpt")
for i in range(num_of_episodes):
model.reset(random_init=True)
rAll = 0
for j in range(1, model.Nsim + 1):
if j % rl.eval_period == 0:
s = rl.state_detection(model.x[j - 1, 1])
a, allQ = sess.run(
[predict, Qout],
feed_dict={inputs: np.identity(len(states))[s:s + 1]})
number = np.random.rand()
if number < 0.0:
a = [np.random.randint(0, len(actions))]