def __init__(self,model,weighted=True,stdDevWeighted=True,toPrune=True): ''' SlopeOneRecommender Class Constructor `model` is the data source model `weighted` is a flag that if it is True, it act as a weighted slope one recommender. `stdDevWeighted` is a flag that if it is True, use standard deviation weighting of diffs `toPrune` is a flag that if it is True, it will prune the irrelevant diffs, represented by one data point. ''' Recommender.__init__(self,model) self.weighted = weighted self.stdDevWeighted = stdDevWeighted self.storage = DiffStorage(self.model,self.stdDevWeighted,toPrune)
def __init__(self,
model,
weighted=True,
stdDevWeighted=True,
toPrune=True):
'''
SlopeOneRecommender Class Constructor
`model` is the data source model
`weighted` is a flag that if it is True, it act as a weighted slope one
recommender.
`stdDevWeighted` is a flag that if it is True, use standard deviation
weighting of diffs
`toPrune` is a flag that if it is True, it will prune the irrelevant
diffs, represented by one data point.
'''
Recommender.__init__(self, model)
self.weighted = weighted
self.stdDevWeighted = stdDevWeighted
self.storage = DiffStorage(self.model, self.stdDevWeighted, toPrune)
class SlopeOneRecommender(Recommender):
"""
A basic "slope one" recommender. This is a recommender specially suitable
when user preferencces are updating frequently as it can incorporate this
information without expensive recomputation. It can also be used as a
weighted slope one recommender.
"""
def __init__(self,
model,
weighted=True,
stdDevWeighted=True,
toPrune=True):
'''
SlopeOneRecommender Class Constructor
`model` is the data source model
`weighted` is a flag that if it is True, it act as a weighted slope one
recommender.
`stdDevWeighted` is a flag that if it is True, use standard deviation
weighting of diffs
`toPrune` is a flag that if it is True, it will prune the irrelevant
diffs, represented by one data point.
'''
Recommender.__init__(self, model)
self.weighted = weighted
self.stdDevWeighted = stdDevWeighted
self.storage = DiffStorage(self.model, self.stdDevWeighted, toPrune)
def recommend(self, userID, howMany, rescore=None):
possibleItemIDs = self.possibleItemIDs(userID)
rec_items = topItems(userID, possibleItemIDs, howMany,
self.estimatePreference, None, None)
return rec_items
def possibleItemIDs(self, userID):
preferences = self.model.ItemIDsFromUser(userID)
recommendableItems = self.storage.recommendableItems()
return [
itemID for itemID in recommendableItems
if itemID not in preferences
]
def estimatePreference(self, **args):
userID = args.get('thingID', None) or args.get('userID', None)
itemID = args.get('itemID', None)
#similarity = args.get('similarity', None)
nHood = args.get('neighborhood', None)
#rescorer = args.get('rescorer', None)
if not nHood:
pref = self.model.PreferenceValue(userID, itemID)
if pref is not None:
return pref
count = 0
totalPreference = 0.0
prefs = self.model.PreferencesFromUser(userID)
averages = self.storage.diffsAverage(userID, itemID, prefs)
for i in range(len(prefs)):
averageDiffValue = averages[i]
if averageDiffValue is not None:
if self.weighted:
weight = self.storage.count(itemID, prefs[i][0])
if self.stdDevWeighted:
stdev = self.storage.standardDeviation(
itemID, prefs[i][0])
if stdev is not None:
weight /= 1.0 + stdev
# If stdev is None, it is because count is 1. Since
# we are weighting by count the weight is already
# low. So we assume stdev is 0.0.
totalPreference += \
weight * (prefs[i][1] + averageDiffValue)
count += weight
else:
totalPreference += prefs[i][1] + averageDiffValue
count += 1
if count <= 0:
return None
# BUGFIX
# itemAverage = self.storage.AverageItemPref(itemID)
# if itemAverage is not None:
# itemAverage = itemAverage.Average()
# return itemAverage
# else:
# return None
else:
return totalPreference / float(count)
class SlopeOneRecommender(Recommender):
"""
A basic "slope one" recommender. This is a recommender specially suitable when user preferencces are
updating frequently as it can incorporate this information without expensive recomputation.
It can also be used as a weighted slope one recommender.
"""
def __init__(self,model,weighted=True,stdDevWeighted=True,toPrune=True):
''' SlopeOneRecommender Class Constructor
`model` is the data source model
`weighted` is a flag that if it is True, it act as a weighted slope one recommender.
`stdDevWeighted` is a flag that if it is True, use standard deviation weighting of diffs
`toPrune` is a flag that if it is True, it will prune the irrelevant diffs, represented by one data point.
'''
Recommender.__init__(self,model)
self.weighted = weighted
self.stdDevWeighted = stdDevWeighted
self.storage = DiffStorage(self.model,self.stdDevWeighted,toPrune)
def recommend(self,userID,howMany,rescore=None):
possibleItemIDs = self.possibleItemIDs(userID)
rec_items = topItems(userID,possibleItemIDs,howMany,self.estimatePreference,None,None)
return rec_items
def possibleItemIDs(self,userID):
preferences = self.model.ItemIDsFromUser(userID)
recommendableItems = self.storage.recommendableItems()
return [ itemID for itemID in recommendableItems if itemID not in preferences]
def estimatePreference(self,**args):
userID = args.get('thingID',None) or args.get('userID',None)
itemID = args.get('itemID',None)
similarity = args.get('similarity',None)
nHood = args.get('neighborhood',None)
rescorer = args.get('rescorer',None)
if not nHood:
pref = self.model.PreferenceValue(userID,itemID)
if pref is not None:
return pref
count = 0
totalPreference = 0.0
prefs = self.model.PreferencesFromUser(userID)
averages = self.storage.diffsAverage(userID,itemID,prefs)
for i in range(len(prefs)):
averageDiffValue = averages[i]
if averageDiffValue is not None:
if self.weighted:
weight = self.storage.count(itemID,prefs[i][0])
if self.stdDevWeighted:
stdev = self.storage.standardDeviation(itemID,prefs[i][0])
if stdev is not None:
weight /= 1.0 + stdev
#If stdev is None, it is because count is 1. Since we are weighting by count
#the weight is already low. So we assume stdev is 0.0.
totalPreference += weight * (prefs[i][1] + averageDiffValue)
count += weight
else:
totalPreference+= prefs[i][1] + averageDiffValue
count+=1
if count <= 0:
return None
#BUGFIX
#itemAverage = self.storage.AverageItemPref(itemID)
#if itemAverage is not None:
# itemAverage = itemAverage.Average()
# return itemAverage
#else:
# return None
else:
return totalPreference/float(count)
