def __init__(self, name, num_states, N):
# create the state and action space
self.inner_size = N
state_space = DiscreteSpace(N)
action_space = DiscreteSpace(2)
# one maps to 2
starting_state = 1
# specify the transition function
transition_func = np.zeros((N, 2), dtype=np.int32)
# iterate over and fill with the transitions
for i in range(N):
transition_func[i, 0] = i - 1
transition_func[i, 1] = i + 1
transition_func[0, 0] = 0
transition_func[N - 1, 1] = N - 1
# now we define the reward function
reward_function = np.zeros((N, 2), dtype=np.float64)
reward_function[0, 0] = 0.001
reward_function[N - 1, 1] = 1
super().__init__(name, num_states, action_space, state_space,
transition_func, reward_function, starting_state)
def __init__(self, name, num_states, N):
# create the state and action space
self.inner_size = N
state_space = DiscreteSpace(2**N - 1)
action_space = DiscreteSpace(3)
# one maps to 2
starting_state = 2**(N - 1)
# specify the transition function
transition_func = np.zeros((2**N - 1, 3), dtype=np.int32)
# iterate over and fill with the transitions
for i in range(2**N - 1):
transition_func[i, 0] = 2 * i + 1
transition_func[i, 1] = int((i - 1) / 2)
transition_func[i, 2] = 2 * i + 2
# set head and leafs
transition_func[0, 1] = 0
for l in range(2**(N - 1) - 1, 2**N - 1):
transition_func[l, 0] = l
transition_func[l, 2] = l
# now we define the reward function
reward_function = np.zeros((2**N - 1, 3), dtype=np.float64)
reward_function[2**N - 2, 0] = N - 1
reward_function[2**N - 2, 2] = N - 1
for l in range(N - 1):
for s in range(2**(l + 1) - 1, 2**(l + 1) + l):
reward_function[l, 1] = -0.1
for l in range(N - 1):
for s in range(2**l - 1, 2**l + l + 1):
reward_function[l, 0] = -0.1
reward_function[l, 2] = -0.1
super().__init__(name, num_states, action_space, state_space,
transition_func, reward_function, starting_state)
def __init__(self, name, num_states, N):
# create the state and action space
self.inner_size = N
state_space = DiscreteSpace(2**N)
action_space = DiscreteSpace(N)
# get size of state and action space
size_space = state_space.get_size()
size_action = action_space.get_size()
# one maps to 2
starting_state = 0
# specify the transition function
transition_func = np.zeros((size_space, size_action), dtype=np.int32)
reward_function = np.zeros((size_space, size_action), dtype=np.float64)
# iterate over and fill with the transitions
for i in range(size_space):
for j in range(size_action):
next_state = self.flip_bit(i, j, N)
transition_func[i, j] = next_state
reward_function[i, j] = np.sign(i - next_state) * np.minimum(
i, next_state)
super().__init__(name, num_states, action_space, state_space,
transition_func, reward_function, starting_state)
def __init__(self, name, num_states, N):
# create the state and action space
self.inner_size = N
state_space = DiscreteSpace(N ** 2)
action_space = DiscreteSpace(4)
# get size of state and action space
size_space = state_space.get_size()
size_action = action_space.get_size()
# one maps to 2
starting_state = 0
# specify the transition function
transition_func = np.zeros((size_space, size_action), dtype=np.int32)
reward_function = np.zeros((size_space, size_action), dtype=np.float64)
# iterate over and fill with the transitions
for i in range(size_space):
transition_func[i, 0] = i if i % N == 0 else i - 1
transition_func[i, 1] = i if i / N < 1 else i - N
transition_func[i, 2] = i if i % N == N - 1 else i + 1
transition_func[i, 3] = i if i / N >= N - 1 else i + N
# now we define the reward function
reward_function[N ** 2 - 1, 2] = 1
reward_function[N ** 2 - 1, 3] = 1
super().__init__(name, num_states, action_space, state_space, transition_func, reward_function, starting_state)
def __init__(self, name, num_states, N):
# create the state and action space
self.inner_size = N
state_space = DiscreteSpace(N**2)
action_space = DiscreteSpace(2)
# get size of state and action space
size_space = state_space.get_size()
size_action = action_space.get_size()
# one maps to 2
starting_state = N + 1
# specify the transition function
transition_func = np.zeros((size_space, size_action), dtype=np.int32)
reward_function = np.zeros((size_space, size_action), dtype=np.float64)
# sample the left action
left = 1
right = 1 - left
chest = 2 * 0 - 1
# iterate over and fill with the transitions
for x in range(N):
for y in range(N):
pos = y * N + x
y = y if y == N - 1 else y + 1
left_x = x if x == 0 else x - 1
right_x = x if x == N - 1 else x + 1
transition_func[pos, left] = y * N + left_x
transition_func[pos, right] = y * N + right_x
reward_function[N**2 - 1, right] = chest
super().__init__(name, num_states, action_space, state_space,
transition_func, reward_function, starting_state)