saG.reverse()
return saG
# Function to return the argmax value from a given dictionary and its key
def max_dict(d):
max_key, max_val = None, float('-inf')
for k, v in d.items():
if v > max_val:
max_val = v
max_key = k
return max_key, max_val
if __name__ == '__main__':
grid = negative_maze(step_cost=-.5)
# Randomly initialize a policy
policy = {}
for s in grid.actions.keys():
policy[s] = np.random.choice(ALL_ACTIONS)
# Q = mean of returns G for state s and action a
Q = {}
G = {}
state = grid.all_states()
for s in state:
if s in grid.actions.keys():
Q[s] = {}
for a in ALL_ACTIONS:
Q[s][a] = 0
return np.random.choice(ALL_ACTIONS)
# Argmax key from a dictionary
def max_dict(d):
max_k, max_v = None, float('-inf')
for k, v in d.items():
if v > max_v:
max_k = k
max_v = v
return max_k, max_v
if __name__ == '__main__':
maze = negative_maze()
print("Values: ")
print_values(maze.rewards, maze)
states = maze.all_states()
# Initialize Q for all states and actions
Q = {}
for s in states:
Q[s] = {}
for a in ALL_ACTIONS:
Q[s][a] = 0
# This would determine the decay in learning rate after each update
state_lr_decay = {}
for s in states:
import numpy as np
from Maze import negative_maze, print_values, print_policy
# Create a minimum threshold to check for convergence
THRESHOLD = 1e-3
# A discount variable is added so that the program is not greedy
GAMMA = 0.9
ALL_ACTIONS = ('U', 'D', 'L', 'R')
if __name__ == '__main__':
# Make a grid object
grid = negative_maze()
print("Rewards: ")
print_values(grid.rewards, grid)
# Randomly initialize a policy for all playable states
policy = {}
states = grid.all_states()
for s in states:
if s in grid.actions.keys():
policy[s] = np.random.choice(ALL_ACTIONS)
print("Random initial Policy")
print_policy(policy, grid)
# Randomly initialize a Value for all playable states and 0 for other states
V = {}
for s in states:
if s in grid.actions.keys():
V[s] = np.random.random()
else: