def get_value(self, state, action):
scaled_state = np.multiply(np.asarray(state).flatten(), self.scale_factor)
tile_indices = asarray(
tiles(self.iht, self.num_tilings, scaled_state),
dtype=int) + (action * self.num_tiles)
return sum(self.theta[tile_indices])
def update(self, state, action, nstep_return):
current_estimate = self.get_value(state, action)
value = nstep_return - current_estimate
scaled_state = np.multiply(
np.asarray(state).flatten(), self.tiling_side_length)
tile_indices = asarray(tiles(self.iht, self.num_tilings, scaled_state),
dtype=int) + (action * self.num_tiles)
self.theta[tile_indices] += self.alpha * value
def get_next_states_values(self, state):
scaled_state = np.multiply(np.asarray(state).flatten(), self.scale_factor)
values = np.zeros(self.num_actions)
for action in range(self.num_actions):
tile_indices = asarray(
tiles(self.iht, self.num_tilings, scaled_state),
dtype=int) + (action * self.num_tiles)
values[action] = np.sum(self.theta[tile_indices])
return values