def hybrDescent(Network, numIter, gamma, beta, delta, eta, eps, nhat, batchSize):
j = 0
mu = 0
lastUpdate = 0
numBatches = Network.m/batchSize
alpha = stepSize(mu, gamma, eta)
while calcAccuracy(Network) < .95 and j < numIter:
j += 1
randvec = range(Network.m)
shuffle(randvec)
oldInputWeight = Network.inputWeight.copy()
oldHiddenWeight = Network.hiddenWeight.copy()
gInput = np.zeros(Network.inputWeight.shape)
hInput = np.zeros(Network.inputWeight.shape)
gHidden = np.zeros(Network.hiddenWeight.shape)
hHidden = np.zeros(Network.hiddenWeight.shape)
for i in range(numBatches-1):
obs = randvec[i*batchSize:(i+1)*batchSize]
prop.calcDeriv(Network, Network.train[obs], obs)
xi = 1/sum(np.power(mu,range(numBatches-i)))
gInput += xi*Network.inputDeriv
gHidden += xi*Network.hiddenDeriv
hInput = mu*hInput + gInput
hHidden = mu*hHidden + gHidden
Network.inputWeight = oldInputWeight - alpha*hInput
Network.hiddenWeight = oldHiddenWeight - alpha*hHidden
lastUpdate += 1
if lastUpdate > nhat or np.linalg.norm(Network.inputWeight-oldInputWeight) < eps:
mu = beta*mu + delta
alpha = stepSize(mu, gamma, eta)
lastUpdate = 0
print "Training set accuracy: %.3f" %calcAccuracy(Network)
def gradDescent(Network, numIter, alpha):
j = 0
while calcAccuracy(Network) < .95 and j < numIter:
j += 1
prop.calcDeriv(Network)
Network.inputWeight -= alpha*Network.inputDeriv
Network.hiddenWeight -= alpha*Network.hiddenDeriv
print "Training set accuracy: %.3f" %calcAccuracy(Network)
def incrDescent(Network, numIter, alpha):
j = 0
while calcAccuracy(Network) < .95 and j < numIter:
j += 1
randvec = range(Network.m)
shuffle(randvec)
for i in randvec:
prop.calcDeriv(Network, Network.train[i], i)
Network.inputWeight -= alpha*Network.inputDeriv
Network.hiddenWeight -= alpha*Network.hiddenDeriv
print "Training set accuracy: %.3f" %calcAccuracy(Network)
def trainNetwork(self, method, *params):
if method == 'GD':
method = descent.gradDescent
elif method == 'ID':
method = descent.incrDescent
else:
method = descent.hybrDescent
method(self, *params)
if self.test is not None:
print "Testing set accuracy"
print utils.calcAccuracy(self, self.test, self.testDigits)
def genPOMDP(filename, reward, cost, gammas, numberOfWorkerPools):
difficulties = getDifficulties(0.1)
numDiffs = len(difficulties)
#Add one absorbing state
numberOfStates = ((numDiffs) * 2) + 1
numberOfActions = numberOfWorkerPools + 2
file = open(filename, 'w')
file.write('discount: 0.9999\n')
file.write('values: reward\n')
file.write('states: %d\n' % numberOfStates)
file.write('actions: %d\n' % numberOfActions)
SUBMITZERO = numberOfWorkerPools
SUBMITONE = numberOfWorkerPools + 1
file.write('observations: Zero One None\n')
for i in range(0, numberOfStates):
for k in range(0, numberOfWorkerPools):
file.write('T: %d : %d : %d %f\n' % (k, i, i, 1.0))
#Add transitions to absorbing state
file.write('T: %d : * : %d %f\n' % (SUBMITZERO, numberOfStates-1, 1.0))
file.write('T: %d : * : %d %f\n' % (SUBMITONE, numberOfStates-1, 1.0))
#Add observations in absorbing state
file.write('O: * : %d : None %f\n' % (numberOfStates-1, 1.0))
for v in range(0, 2):
for diffState in range(0,numDiffs):
#for diffState in product(range(numDiffs), repeat = numberOfWorkerPools):
state = v * numDiffs + diffState
"""for k in range(0, numberOfWorkerPools):
state += (diffState[k] * (numDiffs ** (numberOfWorkerPools - (k+1))))"""
file.write('O: %d: %d : None %f\n' % (SUBMITZERO, state, 1.0))
file.write('O: %d: %d : None %f\n' % (SUBMITONE, state, 1.0))
if v == 0: #if the answer is 0
for k in range(0, numberOfWorkerPools):
file.write('O: %d : %d : Zero %f\n' %
(k, state, calcAccuracy(gammas[k], difficulties[diffState])))
file.write('O: %d : %d : One %f\n' %
(k, state, 1.0 - calcAccuracy(gammas[k], difficulties[diffState])))
else: # if the answer is 1
for k in range(0, numberOfWorkerPools):
file.write('O: %d : %d : Zero %f\n' %
(k, state, 1.0 - calcAccuracy(gammas[k], difficulties[diffState])))
file.write('O: %d : %d : One %f\n' %
(k, state, calcAccuracy(gammas[k], difficulties[diffState])))
for v in range(0,numberOfWorkerPools):
file.write('R: %d : * : * : * %f\n' % (v, -1 * cost[v]))
for i in range(0, numberOfStates-1):
if i < (numberOfStates-1) / 2:
file.write('R: %d : %d : %d : * %f\n' % (SUBMITZERO, i, numberOfStates-1, 1))
file.write('R: %d : %d : %d : * %f\n' % (SUBMITONE, i, numberOfStates-1, reward))
else:
file.write('R: %d : %d : %d : * %f\n' % (SUBMITONE, i, numberOfStates-1, 1))
file.write('R: %d : %d : %d : * %f\n' % (SUBMITZERO, i, numberOfStates-1, reward))
#Add rewards in absorbing state
file.write('R: * : %d : %d : * %f\n' % (numberOfStates-1, numberOfStates-1, 0))
file.close()
