def expected_value_difference(n_states, n_actions, transition_probability,
reward, discount, p_start_state, optimal_value,
true_reward):
"""
Calculate the expected value difference, which is a proxy to how good a
recovered reward function is.
n_states: Number of states. int.
n_actions: Number of actions. int.
transition_probability: NumPy array mapping (state_i, action, state_k) to
the probability of transitioning from state_i to state_k under action.
Shape (N, A, N).
reward: Reward vector mapping state int to reward. Shape (N,).
discount: Discount factor. float.
p_start_state: Probability vector with the ith component as the probability
that the ith state is the start state. Shape (N,).
optimal_value: Value vector for the ground reward with optimal policy.
The ith component is the value of the ith state. Shape (N,).
true_reward: True reward vector. Shape (N,).
-> Expected value difference. float.
"""
policy = value_iteration.find_policy(n_states, n_actions,
transition_probability, reward,
discount)
value = value_iteration.value(policy.argmax(axis=1), n_states,
transition_probability, true_reward,
discount)
evd = optimal_value.dot(p_start_state) - value.dot(p_start_state)
return evd
def expected_value_difference(n_states, n_actions, transition_probability,
reward, discount, p_start_state, optimal_value,
true_reward):
"""
Calculate the expected value difference, which is a proxy to how good a
recovered reward function is.
"""
policy = value_iteration.find_policy(n_states, n_actions,
transition_probability, reward,
discount)
value = value_iteration.value(policy.argmax(axis=1), n_states,
transition_probability, true_reward,
discount)
evd = optimal_value.dot(p_start_state) - value.dot(p_start_state)
return evd