R.SaveBuffer(OUT_DIR+BUFFER_FILE)
print "model saved, replay buffer: ", R.GetOccupency()
L.Save(OUT_DIR+LOG_FILE)
Ws,bs = Q.get_weights()
Q_target.assign(sess, Ws,bs)
save_counter.increment(sess)
saver.save(sess,OUT_DIR+"model.ckpt", global_step=save_counter.evaluate(sess))
R.SaveBuffer(OUT_DIR+BUFFER_FILE)
print "model saved, replay buffer: ", R.GetOccupency()
L.Save(OUT_DIR+LOG_FILE)
sess.close()
sys.exit()
# plot statistics
R_P_l = L.GetLogByName('policy_left')
R_Q_l = L.GetLogByName('network_left')
R_P_m = L.GetLogByName('policy_middle')
R_Q_m = L.GetLogByName('network_middle')
R_P_r = L.GetLogByName('policy_right')
R_Q_r = L.GetLogByName('network_right')
totalR = L.GetLogByName('total_reward')
error = L.GetLogByName('error')
value = L.GetLogByName('estimated_value')
t = np.arange(R_P_l.size)
plt.figure(1)
plt.plot(t, R_Q_l, 'b', t, R_P_l, 'r')
plt.xlabel('Episodes')
plt.ylabel('reward')
plt.title('Puck on the left')
L.Save(OUT_DIR + LOG_FILE)
Ws, bs = Q.get_weights()
Q_target.assign(sess, Ws, bs)
save_counter.increment(sess)
saver.save(sess,
OUT_DIR + "model.ckpt",
global_step=save_counter.evaluate(sess))
R.SaveBuffer(OUT_DIR + BUFFER_FILE)
print "model saved, replay buffer: ", R.GetOccupency()
L.Save(OUT_DIR + LOG_FILE)
sess.close()
# plot statistics
R_P_l = L.GetLogByName('policy_left')
R_Q_l = L.GetLogByName('network_left')
R_P_m = L.GetLogByName('policy_middle')
R_Q_m = L.GetLogByName('network_middle')
R_P_r = L.GetLogByName('policy_right')
R_Q_r = L.GetLogByName('network_right')
t = np.arange(R_P_l.size)
plt.figure(1)
plt.plot(t, R_Q_l, 'b', t, R_P_l, 'r')
plt.figure(2)
plt.plot(t, R_Q_m, 'b', t, R_P_m, 'r')
plt.figure(3)
plt.plot(t, R_Q_r, 'b', t, R_P_r, 'r')
#if UPDATE_TARGET:
# Ws,bs = Q.get_weights()
# Q_targets[cur_target % len(Q_targets)].assign(sess, Ws,bs)
save_counter.increment(sess)
saver.save(sess,
OUT_DIR + "model.ckpt",
global_step=save_counter.evaluate(sess))
R.SaveBuffer(OUT_DIR + BUFFER_FILE)
print "model saved, replay buffer: ", R.GetOccupency()
L.Save(OUT_DIR + LOG_FILE)
sess.close()
# plot statistics
if DISPLAY_STATISTICS:
R_P_l = L.GetLogByName('policy_left')
R_Q_l = L.GetLogByName('network_left')
R_P_m = L.GetLogByName('policy_middle')
R_Q_m = L.GetLogByName('network_middle')
R_P_r = L.GetLogByName('policy_right')
R_Q_r = L.GetLogByName('network_right')
currentR = L.GetLogByName('network_random')
error = L.GetLogByName('error')
value = L.GetLogByName('estimated_value')
t = np.arange(R_P_l.size)
plt.figure()
plt.subplot(231)
plt.plot(t, R_Q_l, 'b', t, R_P_l, 'r')
plt.xlabel('Episodes')
