# pos = env.arraystate2pos(state)
# print(pos)
# optimal_action = np.zeros(2)
# action, optimal_action, a, b, theta = agent.test(env, nb_episodes=500000, visualize=False, nb_max_episode_steps=200, modif = True, pos = pos)
# env.non_random_reset(pos[0], pos[1], pos[2])
# env.render = True
# env.step(action, rand = optimal_action, a = a, b = b, theta = theta)
# state = env.reset()
# pos = env.arraystate2pos(state)
# optimal_action = np.zeros(2)
# optimal_action[0], optimal_action[1], a, b, theta = agent.test(env, nb_episodes=500000, visualize=False, nb_max_episode_steps=200, modif = True, pos = pos)
# env.non_random_reset(pos[0], pos[1], pos[2])
# env.render = True
# env.step(optimal_action, a = a, b = b, theta = theta)
nb_test = 50
for i in range(nb_test):
env.render = False
state = env.reset()
pos = env.arraystate2pos(state)
optimal_action = np.zeros(2)
action, optimal_action, a, b, theta = agent.test(env,
nb_episodes=500000,
visualize=False,
nb_max_episode_steps=200,
modif=True,
pos=pos)
env.non_random_reset(pos[0], pos[1], pos[2])
env.render = True
env.step(action, rand=optimal_action, a=a, b=b, theta=theta)
from carom import Carom
from Constants import *
import numpy as np
from vpython import sleep
def choose_branch(Q, state):
chosen_branch = Q[state].argmax()
return chosen_branch
tree_actions_index = np.load("tree_actions_index.npy")
Q = np.load("treeQmatrix.npy")
env = Carom(render=True)
env.reset()
actions = env.get_actions()
nb_states = 0
index2 = 0
state = 0
episode_reward = 0
nb_branches = len(tree_actions_index[0, 0])
for i in range(tree_actions_index.shape[0]):
nb_states += nb_branches**i
for level in range(tree_actions_index.shape[0]):
chosen_branch = choose_branch(Q, state)
action_index = int(tree_actions_index[level, index2][chosen_branch])
reward, coll_r = env.step3(actions[action_index][0],
actions[action_index][1],
#num_episodes = 2000
#lr = .8
#y = .95
env = Carom(render=False)
#tree_states_index = np.zeros((goal_points,nb_branches**(goal_points-1)))
tree_actions_index = np.zeros((goal_points, nb_branches**(goal_points - 1)),
dtype=myList)
#print(tree_states_index)
actions = env.get_actions()
states_list = [(pos_white.x, pos_white.y, pos_yellow.x, pos_yellow.y,
pos_red.x, pos_red.y)]
#actions =[(0,0,0,20,4),(0,0,0,130,3),(0,0,0,1,8)]
#Q = np.zeros((1, len(actions)))
env.reset()
#env.step(0,0,0,90,5) #a, b, thetha, phi, Vb
#for i in range(num_episodes):
level = 0
state_index = -1
while total_points < sum_goal_points:
level += 1
for i in range(nb_branches**level):
if (i % nb_branches == 0 or i == 0):
state_index += 1
pos_white = vector(states_list[state_index][0],
states_list[state_index][1], 0)
pos_yellow = vector(states_list[state_index][2],
states_list[state_index][3], 0)
pos_red = vector(states_list[state_index][4],
states_list[state_index][5], 0)