return self.sa2x(s, a)
def getPolicyQ(model, s):
# we need Q(s,a) to choose an action
# i.e. a = argmax[a]{ Q(s,a) }
policyQ = {}
for a in allActions:
q = model.predict(s, a)
policyQ[a] = q
return policyQ
if __name__ == '__main__':
grid = negativeGrid(stepCost=-0.1)
# initialize model
model = Model()
# repeat until convergence
t = 1.0
t2 = 1.0
deltas = []
for i in range(20000):
if i % 100 == 0:
t += 10e-3
t2 += 0.01
alpha = learningRate / t2
s = (2, 0)
grid.setState(s)
grid.setState(s)
statesRewardsList.append((s, 0))
#traverse the grid till we reach the terminal state
while not grid.isGameOver():
oldS = s
oldR = r
a = randomAction(policy[s])
r = grid.move(a)
s = grid.getCurrentState()
valueF[oldS] += alpha * (r + gamma * valueF[s] - valueF[oldS])
return valueF
if __name__ == '__main__':
grid = negativeGrid()
valueF = {}
policy = {}
for s in grid.allStates():
valueF[s] = 0
# state -> action
policy = {
(2, 0): 'U',
(1, 0): 'U',
(0, 0): 'R',
(0, 1): 'R',
(0, 2): 'R',
(1, 2): 'R',
(2, 1): 'R',
(2, 2): 'R',
(2, 3): 'U',
import numpy as np
from grid_world import standardGrid, negativeGrid
from iterative_policy_evaluation import printValues, printPolicy
#this basically refers to the windy grid problem, where lets assume a wind is blowing in the cells and tberefore we are not sure that
#our agent would move for sure where it wants to
#the case taken here is that if our agent wants to move to say direction L, it could do so only with a probability of 0.5
#and it could move to any other direction with an equal probability
if __name__ == '__main__':
smallEnough = 10e-4
gamma = 0.9
print('Hola')
grid = negativeGrid(stepCost=-1.0)
states = grid.allStates()
allActions = ('U', 'D', 'L', 'R')
valueF = {}
policy = {}
for s in states:
valueF[s] = 0
# initial policy
for a in grid.actions.keys():
policy[a] = np.random.choice(allActions)
while True:
#policy evaluation step
while True:
largest = 0
for s in states:
if s in policy:
