In [53]:

    

import numpy as np
import pprint
import sys
if "../" not in sys.path:
  sys.path.append("../") 
from lib.envs.gridworld import GridworldEnv


    

In [54]:

    

pp = pprint.PrettyPrinter(indent=2)
env = GridworldEnv()


    

In [55]:

    

def policy_eval(policy, env, discount_factor=1.0, theta=0.00001):
    """
    Evaluate a policy given an environment and a full description of the environment's dynamics.
    
    Args:
        policy: [S, A] shaped matrix representing the policy.
        env: OpenAI env. env.P represents the transition probabilities of the environment.
            env.P[s][a] is a (prob, next_state, reward, done) tuple.
        theta: We stop evaluation once our value function change is less than theta for all states.
        discount_factor: lambda discount factor.
    
    Returns:
        Vector of length env.nS representing the value function.
    """
    # Start with a random (all 0) value function
    V = np.zeros(env.nS)
    while True:
        delta = 0
        # For each state, perform a "full backup"
        for s in range(env.nS):
            v = 0
            # Look at the possible next actions
            for a, action_prob in enumerate(policy[s]):
                # For each action, look at the possible next states...
                for  prob, next_state, reward, done in env.P[s][a]:
                    # Calculate the expected value
                    v += action_prob * prob * (reward + discount_factor * V[next_state])
            # How much our value function changed (across any states)
            delta = max(delta, np.abs(v - V[s]))
            V[s] = v
        # Stop evaluating once our value function change is below a threshold
        if delta < theta:
            break
    return np.array(V)


    

In [56]:

    

random_policy = np.ones([env.nS, env.nA]) / env.nA
v = policy_eval(random_policy, env)


    

In [57]:

    

print("Value Function:")
print(v)
print("")

print("Reshaped Grid Value Function:")
print(v.reshape(env.shape))
print("")


    

    




Value Function:
[  0.         -13.99993529 -19.99990698 -21.99989761 -13.99993529
 -17.9999206  -19.99991379 -19.99991477 -19.99990698 -19.99991379
 -17.99992725 -13.99994569 -21.99989761 -19.99991477 -13.99994569   0.        ]

Reshaped Grid Value Function:
[[  0.         -13.99993529 -19.99990698 -21.99989761]
 [-13.99993529 -17.9999206  -19.99991379 -19.99991477]
 [-19.99990698 -19.99991379 -17.99992725 -13.99994569]
 [-21.99989761 -19.99991477 -13.99994569   0.        ]]

In [51]:

    

# Test: Make sure the evaluated policy is what we expected
expected_v = np.array([0, -14, -20, -22, -14, -18, -20, -20, -20, -20, -18, -14, -22, -20, -14, 0])
np.testing.assert_array_almost_equal(v, expected_v, decimal=2)


    

In [ ]: