def main() :
if len(sys.argv) < 2 :
usage()
method = sys.argv[1]
args = sys.argv[2:]
argDict = {arg.split('=')[0]:arg.split('=')[1] for arg in args}
if method == 'nn' :
trainNN(**argDict)
elif method == 'retrainNN' :
testNN(**argDict)
elif method == 'testNN' :
testNN(loadNN(**argDict))
elif method == 'rl' :
# number of stocks to choose, test set percentage, starting money amount
evalLinUCB(None, None, **argDict)
elif method == 'pg' :
policyGradient(**argDict)
def policyGradient(T=5, maxStaticRewards=10, maxShift=5, maxIterations=1000, alpha=0.1, dataSet='nasdaq100') :
done = False
stockHistory = StockHistory(dataSet)
currentArgs = [5, 5, 5, 0, 1.5, -1.5]
argEpsilons = [1, 1, 1, 0.01, 0.1, 0.1]
CAGRs = evalLinUCB(stockHistory, Featurizer(stockHistory, *currentArgs), alpha=alpha)
currentMax = sum(CAGRs) / len(CAGRs)
print
print 'Initial CAGRs:', CAGRs
print 'Initial max: ', currentMax
f = open('maxArg.txt', 'a')
f.write('Begin again! Excited!!!\n')
f.close()
numStaticRewards = 0
iteration = 0
while numStaticRewards < maxStaticRewards and iteration < maxIterations :
iteration += 1
randomPerturbations = [[randint(-maxShift, maxShift)*epsilon+arg for (arg,epsilon) in zip(currentArgs, argEpsilons)] for t in range(T)]
for perturbation in randomPerturbations :
print 'Trying args:', perturbation
#changes = [[-1 if arg1 < arg2 else 1 if arg1 > arg2 else 0 for (arg1, arg2) in zip(randomArgs, currentArgs)] for randomArgs in randomPerturbations]
featurizers = [Featurizer(stockHistory, *args) for args in randomPerturbations]
results = [evalLinUCB(stockHistory, featurizer, alpha=alpha) for featurizer in featurizers]
"""
less = [[] for i in range(T)]
zero = [[] for i in range(T)]
more = [[] for i in range(T)]
for i in range(T) :
for change in changes[i] :
if change == -1 :
less[i].append(results[i])
elif change == 0 :
zero[i].append(results[i])
else :
more[i].append(results[i])
avgLess = [sum(l)/float(len(l)) for l in less]
avgZero = [sum(z)/float(len(z)) for z in zero]
avgMore = [sum(m)/float(len(m)) for m in more]
for i in range(T) :
l = less[i]
z = zero[i]
m = more[i]
greatestReward = max(l, z, m)
if greatestReward == l :
currentArgs[i] -= argEpsilons[i]
elif greatestReward == m :
currentArgs[i] += argEpsilons[i]
"""
numStaticRewards += 1
for (args, cagrs) in zip(randomPerturbations, results) :
print 'Args: ', args
print 'CAGRs: ', cagrs
result = sum(cagrs) / len(cagrs)
if result > currentMax :
currentMax = result
currentArgs = args
numStaticRewards = 0
resultString = 'Iteration: ' + str(iteration) + ', Static Rewards: ' + str(numStaticRewards) + ', Current Max: ' + str(currentMax) + ', Args: ' + str(currentArgs)
print
print resultString
f = open('maxArg_' + str(alpha) + '.txt', 'a')
f.write(resultString + '\n')
f.close()
#!/usr/local/bin/python
from core.rl.portfolio import evalLinUCB
from core.util.graphics import plot
from core.util.data import Featurizer, StockHistory
alphas = [0.01, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]
testSizes = [0.1*i for i in range(1, 10)]
stockHistory = StockHistory('nasdaq100')
featurizer = Featurizer(stockHistory)
alphaCAGRs = []
for alpha in alphas :
CAGRs = evalLinUCB(stockHistory, featurizer, alpha=alpha)
alphaCAGRs.append(CAGRs)
plot(alphaCAGRs,
xss=[testSizes for i in range(len(alphaCAGRs))],
labels=[str(a) for a in alphas],
xlabel='Testing Percentage',
ylabel='CAGR',
legendLoc='lower right')
