def init_nature(self, naturelabel):
if naturelabel == 'stochastic':
self.nature = Stochastic(self.matchID)
elif naturelabel == 'deterministic':
self.nature = Deterministic(self.matchID)
else:
self.nature = Adversarial(self.matchID)
def adversarial_example(torch_dev_dataset, parameters, ckpt):
print("Start part 3")
# Question 3 - Adversarial example
data_loader = DataLoader(dataset=torch_dev_dataset, batch_size=1)
adversarial_model = SimpleModel()
adversarial_model.load(path=ckpt)
adversarial = Adversarial(adversarial_model, data_loader,
parameters['adversarial_epsilons'],
parameters['path_plots_adversarial'])
adversarial.__attack__()
adversarial.__plot_attack__()
adversarial.__plot_examples__()
print("End part 3")
def __init__(self, N, eta, nature):
self.N = N
self.weights = [1] * self.N
self.eta = eta
self.T = 100
self.matchID = 0
self.h = [0] * self.N
self.naturelabel = nature
self.nature = Stochastic(self.matchID)
self.learner_loss = [0] * self.T
self.expert_loss = [[0 for i in range(self.N)] for j in range(self.T)]
self.sum_learner = 0
self.sum_experts = [0] * self.N
self.init_nature(self.naturelabel)
class WMA(object):
def __init__(self, N, eta, nature):
self.N = N
self.weights = [1] * self.N
self.eta = eta
self.T = 100
self.matchID = 0
self.h = [0] * self.N
self.naturelabel = nature
self.nature = Stochastic(self.matchID)
self.learner_loss = [0] * self.T
self.expert_loss = [[0 for i in range(self.N)] for j in range(self.T)]
self.sum_learner = 0
self.sum_experts = [0] * self.N
self.init_nature(self.naturelabel)
def init_nature(self, naturelabel):
if naturelabel == 'stochastic':
self.nature = Stochastic(self.matchID)
elif naturelabel == 'deterministic':
self.nature = Deterministic(self.matchID)
else:
self.nature = Adversarial(self.matchID)
def makeExpertPrediction(self):
obs = self.nature.generateObservation(self.matchID)
self.h[0] = self.expertOne(obs)
self.h[1] = self.expertTwo(obs)
self.h[2] = self.expertThree(obs)
print "expert vector", self.h
def getLabel(self):
print "nature", self.nature
if self.naturelabel == 'stochastic':
print self.nature.generateLabel(), "is the label"
return self.nature.generateLabel()
if self.naturelabel == 'deterministic':
print self.nature.generateLabel(self.matchID), "is the label"
return self.nature.generateLabel(self.matchID)
else:
print self.nature.generateLabel(self.weights,
self.h), "is the label"
return self.nature.generateLabel(self.weights, self.h)
def makeLearnerPrediction(self):
#return 1 if sum([i*j for (i, j) in zip(self.h, self.weights)])> 0 else -1
summation = 0
for n in range(self.N):
summation += self.h[n] * self.weights[n]
return (summation > 0) - (summation < 0)
def updateWeights(self, trueLabel):
for i in range(len(self.h)):
#print trueLabel
#print "\n"
#print "expert h" + str(self.h[i])
indicator = 1 if (trueLabel != self.h[i]) else 0
#print "indicator is", indicator
#self.weights[i] = [x*(1-(self.eta*indicator)) for x in self.weights]
self.weights[i] = self.weights[i] * (1 - (self.eta * indicator))
print self.weights
self.matchID += 1
def expertOne(self, obs):
return 1
def expertTwo(self, obs):
return -1
def expertThree(self, obs):
if obs % 2 == 0:
return -1
else:
return 1
def plot_loss(self):
plt.plot([expert[0] for expert in self.expert_loss], 'ro',
[expert[1] for expert in self.expert_loss], 'yo',
[expert[2] for expert in self.expert_loss], 'go',
self.learner_loss, 'bo')
plt.xlabel('Time')
plt.ylabel('Losses')
def plot_regret(self, learner_loss, expert_loss):
best_expert_loss = [min(loss) for loss in expert_loss]
regret = [0] * self.T
for t in range(self.T):
regret[t] = (1.0 *
(learner_loss[t] - best_expert_loss[t])) / (t + 1)
plt.figure(2)
plt.plot(regret, 'r-')
plt.xlabel('t')
plt.ylabel('regret')
plt.title(self.nature)
print "plotted regret"
def WeightedMajorityAlgorithm(self):
while self.matchID < self.T:
self.makeExpertPrediction()
trueLabel = self.getLabel()
learnerPrediction = self.makeLearnerPrediction()
self.sum_learner = self.sum_learner + 1 if trueLabel != learnerPrediction else self.sum_learner
self.learner_loss[self.matchID] = self.sum_learner
print "sumlearner is", self.sum_learner
for i in range(self.N):
self.sum_experts[i] = self.sum_experts[
i] + 1 if trueLabel != self.h[i] else self.sum_experts[i]
self.expert_loss[self.matchID][i] = self.sum_experts[i]
#print "ex loss", self.expert_loss
print "sum experts is", self.sum_experts
#indicator = [1 if trueLabel != learnerPrediction else 0]
self.updateWeights(trueLabel)
#print "learner loss is", self.learner_loss
#print "expert loss is", self.expert_loss
self.plot_loss()
self.plot_regret(self.learner_loss, self.expert_loss)
plt.show()