policy_improvement.py

def policy_improvement():
    
    global policy
    
    policy_stable = True
    for i in range(value.shape[0]):
        for j in range(value.shape[1]):
            old_action = policy[i][j]
            
            max_act_val = None
            max_act = None
            
            τ12 = min(i,5)       # if I have say 3 cars at the first location, then I can atmost move 3 from 1 to 2
            τ21 = -min(j,5)      # if I have say 2 cars at the second location, then I can atmost move 2 from 2 to 1
            
            for act in range(τ21,τ12+1):
                σ = expected_reward([i,j], act)
                if max_act_val == None:
                    max_act_val = σ
                    max_act = act
                elif max_act_val < σ:
                    max_act_val = σ
                    max_act = act
                
            policy[i][j] = max_act
            
            if old_action!= policy[i][j]:
                policy_stable = False
    
    return policy_stable