def get_active_tiles(self, state, action):
position = state[0]
velocity = state[1]
max_position, max_velocity = tuple(self.env.observation_space.high)
min_position, min_velocity = tuple(self.env.observation_space.low)
continuous_features = [8*position/(max_position - min_position),
8*velocity/(max_velocity - min_velocity)]
return tiles(self.iht, NUM_OF_TILINGS, continuous_features, [action])
def get_active_tiles(state, action, env):
position = state[0]
velocity = state[1]
max_position, max_velocity = tuple(env.observation_space.high)
min_position, min_velocity = tuple(env.observation_space.low)
max_action = env.action_space.high[0]
min_action = env.action_space.low[0]
return tiles(iht, NUM_OF_TILINGS, [8*position/(max_position - min_position), \
8*velocity/(max_velocity - min_velocity), \
8*action[0]/(max_action - min_action)])
def get_tiles(self, state):
position = state[0]
velocity = state[1]
max_position, max_velocity = tuple(self.env.observation_space.high)
min_position, min_velocity = tuple(self.env.observation_space.low)
continuous_features = [8*position/(max_position - min_position),
8*velocity/(max_velocity - min_velocity)]
tile_nos = tiles(self.iht, NUM_OF_TILINGS, continuous_features)
tile_array = np.zeros((1, MAX_SIZE))
tile_array[:,tile_nos] = 1.0
return tile_array
def get_active_tiles(state, action, env):
position = state[0]
velocity = state[1]
x3 = state[2]
x4 = state[3]
max_position, max_velocity, max_x3, max_x4 = tuple(env.observation_space.high)
min_position, min_velocity, min_x3, min_x4 = tuple(env.observation_space.low)
return tiles(iht, NUM_OF_TILINGS, [8*position/(max_position - min_position),
8*velocity/(max_velocity - min_velocity),
8*x3/(max_x3 - min_x3),
8*x4/(max_x4 - min_x4)], [action])
def get_active_tiles_tensor(self, state, action):
position = state[0]
velocity = state[1]
acceleration = state[1] - self.old_velocity
max_position, max_velocity = tuple(self.env.observation_space.high)
min_position, min_velocity = tuple(self.env.observation_space.low)
max_acceleration = max_velocity - min_velocity
continuous_features = [8*position/(max_position - min_position),
8*velocity/(max_velocity - min_velocity),
8*acceleration/(2*max_acceleration)]
tile_nos = tiles(self.iht, NUM_OF_TILINGS, continuous_features, [action])
tile_array = np.zeros((MAX_SIZE, 1))
tile_array[tile_nos] = 1.0
return tile_array
def get_active_tiles_tensor(self, state, action):
position = state[0]
velocity = state[1]
acceleration = state[1] - self.old_velocity
max_position, max_velocity = tuple(self.env.observation_space.high)
min_position, min_velocity = tuple(self.env.observation_space.low)
max_acceleration = max_velocity - min_velocity
continuous_features = [8*position/(max_position - min_position),
8*velocity/(max_velocity - min_velocity),
8*acceleration/(2*max_acceleration)]
tile_nos = tiles(self.iht, NUM_OF_TILINGS, continuous_features, [action])
tile_array = np.zeros((MAX_SIZE, 1))
tile_array[tile_nos] = 1.0
# sparse_tensor = tf.SparseTensor(indices = map(lambda x: [0, x], tile_nos),
# values = [1.0]*len(tile_nos),
# shape = [1,MAX_SIZE])
return tile_array