from RewardMatrix import *
from Training import *
from Live import *
from MTM import *
from Ranking import *
from QMatrix import *
from PerformanceMeasures import *
from Plots import *
import calendar
if __name__ == "__main__":
dbObject = DBUtils()
rankingObject = Ranking()
mtmObject = MTM()
rewardMatrixObject = RewardMatrix()
qMatrixObject = QMatrix()
trainingObject = Training()
liveObject = Live()
reallocationObject = Reallocation()
plotObject = Plots()
performanceObject = PerformanceMeasures()
dbObject.dbConnect()
'''
dbObject.dbQuery("DELETE FROM asset_allocation_table")
dbObject.dbQuery("DELETE FROM asset_daily_allocation_table")
dbObject.dbQuery("DELETE FROM mtm_table")
dbObject.dbQuery("DELETE FROM tradesheet_data_table")
dbObject.dbQuery("DELETE FROM reallocation_table")
# Function to check similarity in 2 matrices based on squared difference between corresponding values
def checkSimilarityMatrices(self, m1, m2):
diff = 0
for i in range(0, 3, 1):
for j in range(0, 3, 1):
diff = diff + (m1[i, j] - m2[i, j])**2
if diff < 0.01:
return True
else:
return False
if __name__ == "__main__":
dbObject = DBUtils()
dbObject.dbConnect()
rewardMatrixObject = RewardMatrix()
qMatrixObject = QMatrix()
startDate = 20120409
startTime = timedelta(hours=10, minutes=30)
endDate = 20120409
endTime = timedelta(hours=11, minutes=30)
resultIndividuals = dbObject.getIndividuals(startDate, startTime, endDate,
endTime)
for individualId, dummy in resultIndividuals:
rewardMatrix = rewardMatrixObject.computeRM(individualId, startDate,
startTime, endDate,
endTime, dbObject)
qMatrixObject.calculateQMatrix(rewardMatrix, individualId, dbObject)
break
def feedback(self, alpha, gamma, beta, individualFactor, zeroRange, greedyLevel, workId):
logging.basicConfig(filename=gv.logFileName + str(workId) + '.log', level=logging.INFO, format='%(asctime)s %(message)s')
print(str(datetime.now()) + " Starting Q Learning for : ")
print("alpha = " + str(alpha))
print("gamma = " + str(gamma))
print("beta = " + str(beta))
print("individual factor = " + str(individualFactor))
print("zero range = " + str(zeroRange))
print("greedy level = " + str(greedyLevel))
logging.info("Starting Q Learning for : ")
logging.info("alpha = " + str(alpha))
logging.info("gamma = " + str(gamma))
logging.info("beta = " + str(beta))
logging.info("individual factor = " + str(individualFactor))
logging.info("zero range = " + str(zeroRange))
logging.info("greedy level = " + str(greedyLevel))
logging.info("\n")
setupObject = Setup(alpha, gamma, beta, individualFactor, zeroRange, greedyLevel)
[variableString, latestIndividualTable, trainingTradesheetTable, trainingAssetTable, qMatrixTable, reallocationTable, assetTable, dailyAssetTable, newTradesheetTable] = setupObject.createQLearningTables()
dbObject = DBUtils(alpha, gamma, beta, individualFactor, zeroRange, greedyLevel, latestIndividualTable, trainingTradesheetTable, trainingAssetTable, qMatrixTable, reallocationTable, assetTable, dailyAssetTable, newTradesheetTable)
mtmObject = NetPL_Trades()
rewardMatrixObject = RewardMatrix(alpha)
qMatrixObject = QMatrix(gamma, greedyLevel, beta)
trainingObject = Training()
liveObject = Live()
reallocationObject = Reallocation()
performanceObject = PerformanceMeasures()
performanceOutfileName = gv.performanceOutfileNameBase + variableString + '.csv'
performanceMonthlyOutfileName = gv.performanceMonthlyOutfileNameBase + variableString + '.csv'
dbObject.dbConnect()
dbObject.dbQuery("DELETE FROM " + assetTable)
dbObject.dbQuery("DELETE FROM " + dailyAssetTable)
dbObject.dbQuery("DELETE FROM " + newTradesheetTable)
dbObject.dbQuery("DELETE FROM " + reallocationTable)
dbObject.dbQuery("DELETE FROM " + qMatrixTable)
dbObject.dbQuery("DELETE FROM " + trainingAssetTable)
dbObject.dbQuery("DELETE FROM " + trainingTradesheetTable)
dbObject.dbQuery("DELETE FROM " + latestIndividualTable)
rankingStartDate = gv.startDate
rankingEndDate = rankingStartDate + timedelta(days=gv.rankingDays)
trainingStartDate = rankingEndDate + timedelta(days=1)
trainingEndDate = trainingStartDate + timedelta(days=gv.initializationDays)
liveStartDate = trainingEndDate + timedelta(days=1)
testingStartDate = liveStartDate
liveEndDate = liveStartDate + timedelta(days=gv.liveDays)
testingEndDate = liveEndDate
periodEndDate = gv.endDate
startTime = timedelta(hours=9, minutes=15)
walkforward = 1
dbObject.resetAssetAllocation(liveStartDate, startTime)
done = False
while (not done):
dbObject.resetAssetTraining()
trainingObject.train(trainingStartDate, trainingEndDate, walkforward, dbObject, mtmObject, rewardMatrixObject, qMatrixObject)
dbObject.resetLatestIndividualsWalkForward()
liveObject.live(liveStartDate, liveEndDate, walkforward, dbObject, mtmObject, rewardMatrixObject, qMatrixObject, reallocationObject)
walkforward += 1
if liveEndDate>=periodEndDate:
done = True
else:
dbObject.updateQMatrixTableWalkForward()
dbObject.updateAssetWalkForward()
trainingEndDate = liveEndDate
trainingStartDate = trainingEndDate - timedelta(days=gv.initializationDays)
liveStartDate = liveEndDate + timedelta(days=1)
liveEndDate = liveStartDate + timedelta(days=gv.liveDays)
if liveEndDate>periodEndDate:
liveEndDate = periodEndDate
with open(performanceOutfileName, 'w') as fp:
w = csv.writer(fp)
w.writerow(["original performance", "number of trades", "q learning performance", "number of trades"])
[performanceRef, tradesRef] = performanceObject.CalculateReferenceTradesheetPerformanceMeasures(testingStartDate, periodEndDate, dbObject)
[performance, trades] = performanceObject.CalculateTradesheetPerformanceMeasures(testingStartDate, periodEndDate, dbObject)
[performanceTraining, tradesTraining] = performanceObject.CalculateTrainingTradesheetPerformanceMeasures(testingStartDate, periodEndDate, dbObject)
w.writerow([performanceRef, tradesRef, performance, trades, performanceTraining, tradesTraining])
done = False
with open(performanceMonthlyOutfileName, 'w') as fp:
w = csv.writer(fp)
w.writerow(["original performance", "number of trades", "q learning performance", "number of trades", "q learning training performance", "number of trades"])
#w.writerow(["q learning performance", "number of trades"])
while not done:
[performanceRef, tradesRef] = performanceObject.CalculateReferenceTradesheetPerformanceMeasures(testingStartDate, testingEndDate, dbObject)
[performance, trades] = performanceObject.CalculateTradesheetPerformanceMeasures(testingStartDate, testingEndDate, dbObject)
[performanceTraining, tradesTraining] = performanceObject.CalculateTrainingTradesheetPerformanceMeasures(testingStartDate, testingEndDate, dbObject)
w.writerow([performanceRef, tradesRef, performance, trades, performanceTraining, tradesTraining])
#w.writerow([performance, trades])
if testingEndDate>=periodEndDate:
done = True
else:
testingStartDate = testingEndDate + timedelta(days=1)
testingEndDate = testingStartDate + timedelta(days=gv.liveDays)
if testingEndDate>periodEndDate:
testingEndDate = periodEndDate
dbObject.dbClose()