Implementing value iteration
Value iteration is a key method in RL for finding the optimal policy. It iteratively improves the value function for each state until it converges, resulting in the discovery of the optimal policy. You'll start with an initialized value function V and policy, both preloaded for you. Then, you'll update them in a loop until the value function converges and see the policy in action.
The get_max_action_and_value(state, V) function has been pre-loaded for you.
Diese Übung ist Teil des Kurses
Reinforcement Learning with Gymnasium in Python
Anleitung zur Übung
- For each state, find the action with the maximum Q-value (
max_action) and its corresponding value (max_q_value). - Update the
new_Vdictionary and thepolicybased onmax_actionandmax_q_value. - Check for convergence by checking if the difference between
new_vandVfor every state is less thanthreshold.
Interaktive Übung
Vervollständige den Beispielcode, um diese Übung erfolgreich abzuschließen.
threshold = 0.001
while True:
new_V = {}
for state in range(num_states-1):
# Get action with maximum Q-value and its value
max_action, max_q_value = ____
# Update the value function and policy
new_V[state] = ____
policy[state] = ____
# Test if change in state values is negligeable
if all(abs(____ - ____) < ____ for state in ____):
break
V = new_V
render_policy(policy)