def predictLot(filename, time, AvailNum_weightsVector, Price_weightsVector, lotid, lotLocation, final_dest):
'''
Predicts the availNum and price at |time| by averaging the prediction result from (time-timeBuffer) to (time)
Returns a tuple of the prediction result
'''
fp = open("../data"+filename, 'r')
HourMin = time[0]*60+time[1]
timeBuffer = 10 # the time (in minutes), ahead of arrival_time, that we will consider when we do prediction
count = 0
availNumEstimate = 0.0
priceEstimate = 0.0
for line in fp:
parsedLine = util.exactSingleLineFromFine(line)
_, currHour, currMin = util.convertTimeStampToDate(parsedLine[0])
currHourMin = currHour*60+currMin
if HourMin - currHourMin <= timeBuffer:
count += 1
phi, availNum, price = model.extractRecordFeatures(line,util.locDict, util.eventDict)
if len(phi)<=0 or availNum < 0 or price < 0:
continue
availNumEstimate += round(util.sparseVectorDotProduct(AvailNum_weightsVector, phi))
priceEstimate += util.sparseVectorDotProduct(Price_weightsVector, phi)
# averaging
availNumEstimate /= count
priceEstimate /= count
dist = util.calculateDistance(final_dest, lotLocation) # distance between lot and final location
return (lotid, lotLocation, round(availNumEstimate), priceEstimate, dist)
def readFileUpdateWeight(AvailNum_weightsVector, Price_weightsVector, filepath = 'NA', locDict='NA', eventDict='NA', alpha=.5, eta_0=1.):
'''
Reads each line in the filepath and update weightsVector (a Counter obj representing
sparse vector) by stochastic gradient descent, for predicting both AvailNum and Price
w[i] = w[i] + eta*(y[i]-dotProd(w,phi))*phi[i]
update learning rate by eta = eta_0/t^alpha
'''
# _AvailNum_weightsVector = AvailNum_weightsVector.copy()
# _Price_weightsVector = Price_weightsVector.copy()
if not os.path.exists(filepath):
raise "File doesn't exist!!"
return _weightsVector
else:
fp = open(filepath, 'r')
for t, line in enumerate(fp):
eta = eta_0/(t+1)**alpha
phi, availNum, currPrice = model.extractRecordFeatures(line, locDict, eventDict)
if len(phi) <= 0 or availNum < 0 or currPrice < 0: # if nothing changed for this line
continue
AvailNum_dotProd = util.sparseVectorDotProduct(phi, AvailNum_weightsVector)
Price_dotProd = util.sparseVectorDotProduct(phi, Price_weightsVector)
for key in phi:
# print "-------",y-dotProd
AvailNum_weightsVector[key] += eta * (availNum - AvailNum_dotProd) * phi[key]
Price_weightsVector[key] += eta * (currPrice - Price_dotProd) * phi[key]
fp.close()
return (AvailNum_weightsVector, Price_weightsVector)
def predictLot(filename, time, AvailNum_weightsVector, Price_weightsVector,
lotid, lotLocation, final_dest):
'''
Predicts the availNum and price at |time| by averaging the prediction result from (time-timeBuffer) to (time)
Returns a tuple of the prediction result
'''
fp = open("../data" + filename, 'r')
HourMin = time[0] * 60 + time[1]
timeBuffer = 10 # the time (in minutes), ahead of arrival_time, that we will consider when we do prediction
count = 0
availNumEstimate = 0.0
priceEstimate = 0.0
for line in fp:
parsedLine = util.exactSingleLineFromFine(line)
_, currHour, currMin = util.convertTimeStampToDate(parsedLine[0])
currHourMin = currHour * 60 + currMin
if HourMin - currHourMin <= timeBuffer:
count += 1
phi, availNum, price = model.extractRecordFeatures(
line, util.locDict, util.eventDict)
if len(phi) <= 0 or availNum < 0 or price < 0:
continue
availNumEstimate += round(
util.sparseVectorDotProduct(AvailNum_weightsVector, phi))
priceEstimate += util.sparseVectorDotProduct(
Price_weightsVector, phi)
# averaging
availNumEstimate /= count
priceEstimate /= count
dist = util.calculateDistance(
final_dest, lotLocation) # distance between lot and final location
return (lotid, lotLocation, round(availNumEstimate), priceEstimate, dist)
def computeError(weights, y, count, sumErr, Yvec, estimateVec, rounding=1):
estimate = util.sparseVectorDotProduct(weights, phi)
if rounding:
estimate = round(estimate)
if abs(y - estimate) < 1e-3:
sumErr += 0
count += 1
elif y > 0:
count += 1
sumErr += abs(y - estimate) / float(y)
Yvec.append(y)
estimateVec.append(estimate)
return (sumErr, count)
def computeError(weights, y, count, sumErr, Yvec, estimateVec,rounding=1):
estimate = util.sparseVectorDotProduct(weights, phi)
if rounding:
estimate = round(estimate)
if abs(y-estimate) < 1e-3:
sumErr += 0
count += 1
elif y > 0:
count +=1
sumErr += abs(y-estimate) / float(y)
Yvec.append(y)
estimateVec.append(estimate)
return (sumErr, count)
def test2(filename, locDict, eventDict, weightsVector, plotting=0):
'''
Old testing function, for just testing AvailNum
'''
fp = open("../data" + filename, 'r')
count = 0
sumErr = 0
Yvec = list()
estimateVec = list()
for line in fp:
#print line
phi, y, _ = model.extractRecordFeatures(line, locDict, eventDict)
if len(phi) <= 0 or y < 0:
continue
estimate = util.sparseVectorDotProduct(weightsVector, phi)
estimate = round(estimate)
# print "==========show feature vector==========",phi
# print "real", y, "est:", estimate, "diff error", y-estimate
if abs(y - estimate) < 1e-3:
# print "error rate", 0
sumErr += 0
count += 1
elif y > 0:
# print "error rate", (y-estimate)/y
count += 1
sumErr += abs(y - estimate) / y
Yvec.append(y)
estimateVec.append(estimate)
fp.close()
avgErr = sumErr / count
print "--average error rate--", avgErr
def test2(filename, locDict, eventDict, weightsVector, plotting=0):
'''
Old testing function, for just testing AvailNum
'''
fp = open("../data"+filename, 'r')
count = 0
sumErr = 0
Yvec = list()
estimateVec = list()
for line in fp:
#print line
phi, y, _ = model.extractRecordFeatures(line,locDict, eventDict)
if len(phi)<=0 or y < 0:
continue
estimate = util.sparseVectorDotProduct(weightsVector, phi)
estimate = round(estimate)
# print "==========show feature vector==========",phi
# print "real", y, "est:", estimate, "diff error", y-estimate
if abs(y-estimate) < 1e-3:
# print "error rate", 0
sumErr += 0
count += 1
elif y > 0:
# print "error rate", (y-estimate)/y
count +=1
sumErr += abs(y-estimate) / y
Yvec.append(y)
estimateVec.append(estimate)
fp.close()
avgErr = sumErr/count
print "--average error rate--", avgErr
