def splitColdStart(reviews,minRatings):
outputSplitter = []
userBizSplit = []
userOnlySplit = []
bizOnlySplit = []
coldSplit = []
# List of 4 options:
# coldStartCalc[0] = both user and biz exist
# coldStartCalc[1] = only user exists
# coldStartCalc[2] = only business exists
# coldStartCalc[3] = neither user nor business exists
coldStartCalc = [0,0,0,0]
bizDict = YelpPredictor.getAverageBusinessStars()
userDict = YelpPredictor.getAverageUserStars()
# get list of training reviews
# so that we can calculate
#reviews = getUserBusinessRatings()
for entry in reviews:
user_id = entry[0]
business_id = entry[1]
# if user exists
if userDict.has_key(user_id):
userRevAry = userDict.get(user_id)
# if user meets review threshold
if userRevAry[1] >= minRatings:
# if business exists
if bizDict.has_key(business_id):
bizRevAry = bizDict.get(business_id)
# if business meets review threshold
if bizRevAry[1] >= minRatings:
coldStartCalc[0]+=1
userBizSplit.append(entry)
else:
coldStartCalc[1]+=1
userOnlySplit.append(entry)
# End IF/ELSE business and user reviews meet the threshold
# End IF user and business exist
else:
coldStartCalc[1]+=1
userOnlySplit.append(entry)
# End IF/ELSE business exists given that the user exists
# End IF user reviews meet the threshold
else:
# user reviews do not meet threshold
# if business exists
if bizDict.has_key(business_id):
bizRevAry = bizDict.get(business_id)
# if business meets review threshold
if bizRevAry[1] >= minRatings:
coldStartCalc[2]+=1
bizOnlySplit.append(entry)
else:
coldStartCalc[3]+=1
coldSplit.append(entry)
# End IF/ELSE business reviews meet threshold, user reviews do not
# End IF user and business exist
else:
coldStartCalc[3]+=1
coldSplit.append(entry)
# End IF/ELSE business exists given that the user review threshold not met
# End IF/ELSE user review threshold met
# End IF user exists
else:
# if business exists, but user doesn't
if bizDict.has_key(business_id):
bizRevAry = bizDict.get(business_id)
# check for business review threshold
if bizRevAry[1] >= minRatings:
coldStartCalc[2]+=1
bizOnlySplit.append(entry)
else:
coldStartCalc[3]+=1
coldSplit.append(entry)
# End IF/ELSE business review threshold met, if user doesn't exist
# End IF business exists, but user doesn't
else:
coldStartCalc[3]+=1
coldSplit.append(entry)
# End IF/ELSE business exists when user doesn't
# End IF/ELSE user exists
# End FOR each user in training review set
print 'both user and biz: '+str(coldStartCalc[0])+', '+str(float(coldStartCalc[0]) / float(sum(coldStartCalc)))
print 'only user: '******', '+str(float(coldStartCalc[1]) / float(sum(coldStartCalc)))
print 'only biz: '+str(coldStartCalc[2])+', '+str(float(coldStartCalc[2]) / float(sum(coldStartCalc)))
print 'neither: '+str(coldStartCalc[3])+', '+str(float(coldStartCalc[3]) / float(sum(coldStartCalc)))
outputSplitter.append(userBizSplit)
outputSplitter.append(userOnlySplit)
outputSplitter.append(bizOnlySplit)
outputSplitter.append(coldSplit)
print str(len(reviews))+' input reviews'
print str(len(userBizSplit))+' userBizSplit'
print str(len(userOnlySplit))+' userOnlySplit'
print str(len(bizOnlySplit))+' bizOnlySplit'
print str(len(coldSplit))+' coldSplit'
return outputSplitter
def simulateKaggleMix(reviews,minRatings):
outputSplitter = []
userBizSplit = []
userOnlySplit = []
bizOnlySplit = []
coldSplit = []
# List of 4 options:
# coldStartCalc[0] = both user and biz exist
# coldStartCalc[1] = only user exists
# coldStartCalc[2] = only business exists
# coldStartCalc[3] = neither user nor business exists
coldStartCalc = [0,0,0,0]
# Calculate target #s for each scenario option.
# First, get total number of reviews to process.
totalReviews = len(reviews)
# Then calculate target #s for each scenario
# based on analysis of final Kaggle test set
targetCalc = [math.trunc(0.33*totalReviews),math.trunc(0.11*totalReviews),math.trunc(0.41*totalReviews),math.trunc(0.15*totalReviews)]
# Create placeholder to accumulate reviews exceeding target numbers.
# This will be used to re-balance after all true distributions has been completed.
rebalance = []
bizDict = YelpPredictor.getAverageBusinessStars()
userDict = YelpPredictor.getAverageUserStars()
# get list of training reviews
# so that we can calculate
#reviews = getUserBusinessRatings()
for entry in reviews:
user_id = entry[0]
business_id = entry[1]
# if user exists
if userDict.has_key(user_id):
userRevAry = userDict.get(user_id)
# if user meets review threshold
if userRevAry[1] >= minRatings:
# if business exists
if bizDict.has_key(business_id):
bizRevAry = bizDict.get(business_id)
# if business meets review threshold
if bizRevAry[1] >= minRatings:
# if we have not reached the target number of reviews for user+biz scenario
if (coldStartCalc[0] < targetCalc[0]):
coldStartCalc[0]+=1
userBizSplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE user+biz target reviews met
else:
# if we have not reached the target number of reviews for userOnly scenario
if (coldStartCalc[1] < targetCalc[1]):
coldStartCalc[1]+=1
userOnlySplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE userOnly target reviews met
# End IF/ELSE business and user reviews meet the threshold
# End IF user and business exist
else:
# if we have not reached the target number of reviews for userOnly scenario
if (coldStartCalc[1] < targetCalc[1]):
coldStartCalc[1]+=1
userOnlySplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE userOnly target reviews met
# End IF/ELSE business exists given that the user exists
# End IF user reviews meet the threshold
else:
# user reviews do not meet threshold
# if business exists
if bizDict.has_key(business_id):
bizRevAry = bizDict.get(business_id)
# if business meets review threshold
if bizRevAry[1] >= minRatings:
# if we have not reached the target number of reviews for bizOnly scenario
if (coldStartCalc[2] < targetCalc[2]):
coldStartCalc[2]+=1
bizOnlySplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE bizOnly target reviews met
else:
# if we have not reached the target number of reviews for trueCold scenario
if (coldStartCalc[3] < targetCalc[3]):
coldStartCalc[3]+=1
coldSplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE trueCold target reviews met
# End IF/ELSE business reviews meet threshold, user reviews do not
# End IF user and business exist
else:
# if we have not reached the target number of reviews for trueCold scenario
if (coldStartCalc[3] < targetCalc[3]):
coldStartCalc[3]+=1
coldSplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE trueCold target reviews met
# End IF/ELSE business exists given that the user review threshold not met
# End IF/ELSE user review threshold met
# End IF user exists
else:
# if business exists, but user doesn't
if bizDict.has_key(business_id):
bizRevAry = bizDict.get(business_id)
# check for business review threshold
if bizRevAry[1] >= minRatings:
# if we have not reached the target number of reviews for bizOnly scenario
if (coldStartCalc[2] < targetCalc[2]):
coldStartCalc[2]+=1
bizOnlySplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE bizOnly target reviews met
else:
# if we have not reached the target number of reviews for trueCold scenario
if (coldStartCalc[3] < targetCalc[3]):
coldStartCalc[3]+=1
coldSplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE trueCold target reviews met
# End IF/ELSE business review threshold met, if user doesn't exist
# End IF business exists, but user doesn't
else:
# if we have not reached the target number of reviews for trueCold scenario
if (coldStartCalc[3] < targetCalc[3]):
coldStartCalc[3]+=1
coldSplit.append(entry)
else:
rebalance.append(entry)
# End IF/ELSE trueCold target reviews met
# End IF/ELSE business exists when user doesn't
# End IF/ELSE user exists
# End FOR each user in training review set
# rebalance any scenarios that were not naturally completed
# userOnly
while ((coldStartCalc[1] < targetCalc[1]) and (len(rebalance) > 0)):
coldStartCalc[1]+=1
userOnlySplit.append(rebalance.pop())
# End WHILE userOnly is unbalanced and we have available rebalance stock to use
# bizOnly
while ((coldStartCalc[2] < targetCalc[2]) and (len(rebalance) > 0)):
coldStartCalc[2]+=1
bizOnlySplit.append(rebalance.pop())
# End WHILE bizOnly is unbalanced and we have available rebalance stock to use
# true cold start
while ((coldStartCalc[3] < targetCalc[3]) and (len(rebalance) > 0)):
coldStartCalc[3]+=1
coldSplit.append(rebalance.pop())
# End WHILE userOnly is unbalanced and we have available rebalance stock to use
print 'both user and biz: '+str(coldStartCalc[0])+', '+str(float(coldStartCalc[0]) / float(sum(coldStartCalc)))+'; Target: '+str(targetCalc[0])+', 33%'
print 'only user: '******', '+str(float(coldStartCalc[1]) / float(sum(coldStartCalc)))+'; Target: '+str(targetCalc[1])+', 11%'
print 'only biz: '+str(coldStartCalc[2])+', '+str(float(coldStartCalc[2]) / float(sum(coldStartCalc)))+'; Target: '+str(targetCalc[2])+', 41%'
print 'neither: '+str(coldStartCalc[3])+', '+str(float(coldStartCalc[3]) / float(sum(coldStartCalc)))+'; Target: '+str(targetCalc[3])+', 15%'
outputSplitter.append(userBizSplit)
outputSplitter.append(userOnlySplit)
outputSplitter.append(bizOnlySplit)
outputSplitter.append(coldSplit)
print str(len(reviews))+' input reviews'
print str(len(userBizSplit))+' userBizSplit'
print str(len(userOnlySplit))+' userOnlySplit'
print str(len(bizOnlySplit))+' bizOnlySplit'
print str(len(coldSplit))+' coldSplit'
return outputSplitter
