def find_categories_best_k(k):
Steps = 5000
Alpha = 0.05
#arrID2... are maps from the array indicies to the business and user id
P, Q, arrID2bid, arrID2uid = get_p_q_best(k, Steps, Alpha)
zipQ = zip(*Q)
latentNum = 0
for l in zipQ: # {
maxVal = max(l)
cutOff = 0.1 * maxVal
print " Cutoff is: " + str(cutOff)
relevantBus = []
for i in xrange(0, len(l)):
if l[i] > cutOff:
# this business is relevant to this latent variable
# save the id (index is id-1) to use in db look up later
for elemI in xrange(0, len(relevantBus)):
if relevantBus[elemI][1] < l[i]:
relevantBus.insert(elemI, [arrID2bid[i], l[i]])
if len(relevantBus) == 0:
relevantBus.append([arrID2bid[i], l[i]])
otherList = []
for r in relevantBus:
print(r[1])
otherList.append(r[0])
print("\n")
print("\n")
print("\n")
#print " " + str(len(relevantBus)) + " businesses past cutoff"
# For this latent variable, we now have all businesses IDs,
# print out all of the labels associated with these businesses
fp = open(settings.RESULTS_DIR + "latent_" + str(latentNum), 'w')
buses = Business.objects.filter(pk__in=otherList)
for b in buses:
keywords = b.keywords.all()
#print " " + str(len(keywords)) + " keywords for business"
nm = b.name
fp.write(nm.encode("utf8") + "\n")
fp.write(b.address.encode("utf8") + "\n")
for k in keywords:
fp.write(str(k) + "\n")
fp.write("\n")
fp.close()
latentNum = latentNum + 1
def build_pred_server():
k = 42
Steps = 5000
Alpha = 0.05
print("BEFORE")
P, Q, arrID2bid, arrID2uid = get_p_q_best(k, Steps, Alpha)
print("AFTER)")
# print(len(arrID2uid))
# print(len(P))
# #print(len(P))
i = 0
factors = []
for row in P:
k = 0
actualUID = arrID2uid[i]
#this user hasn't rated anything
if actualUID == 0:
continue
usr = User.objects.get(id=actualUID)
for col in row:
uf = UserFactor(user=usr, latentFactor=k, relation=col)
factors.append(uf)
k += 1
i += 1
UserFactor.objects.bulk_create(factors)
i = 0
factors = []
for row in Q:
actualBID = arrID2bid[i]
#this business hasn't been rated
if actualBID == 0:
continue
bus = Business.objects.get(id=actualBID)
k = 0
for col in row:
bf = BusinessFactor(business=bus, latentFactor=k, relation=col)
factors.append(bf)
k += 1
i += 1
BusinessFactor.objects.bulk_create(factors)
Predictions = numpy.dot(P,numpy.transpose(Q))
i = 0
predictions = []
for row in Predictions:
print(len(row))
j = 0
bus = Business.objects.get(id=arrID2bid[j])
for cell in row:
usr = User.objects.get(id=arrID2uid[i])
p = Recommendation(business=bus,recommendation=cell,username=usr)
predictions.append(p)
j+=1
i+=1
Recommendation.objects.bulk_create(predictions)
transaction.commit();
def build_predictions():
k = 42
Steps = 5000
Alpha = 0.05
print("Run Matrix Factorization to get predictions")
print("Alpha: "+str(Alpha)+"\n")
print("Steps: "+str(Steps)+"\n")
print("K: "+str(k)+"\n")
P, Q, arrID2bid, arrID2uid = get_p_q_best(k, Steps, Alpha)
print("Done running Matrix Factorization")
i = 0
factors = []
print("Creating user factors and saving them")
for row in P:
k = 0
actualUID = arrID2uid[i]
#this user hasn't rated anything
if actualUID == 0:
continue
usr = User.objects.get(id=actualUID)
for col in row:
uf = UserFactor(user=usr, latentFactor=k, relation=col)
factors.append(uf)
k += 1
i += 1
UserFactor.objects.bulk_create(factors)
i = 0
factors = []
print("Creating business factors and saving them")
for row in Q:
actualBID = arrID2bid[i]
#this business hasn't been rated
if actualBID == 0:
continue
bus = Business.objects.get(id=actualBID)
k = 0
for col in row:
bf = BusinessFactor(business=bus, latentFactor=k, relation=col)
factors.append(bf)
k += 1
i += 1
BusinessFactor.objects.bulk_create(factors)
print(factors)
Predictions = numpy.dot(P,numpy.transpose(Q))
if Predictions == 0:
logger.debug('Predictions == 0. Returning')
print('Predictions == 0. Returning')
return
print(Predictions)
i = 0
predictions = []
for row in Predictions:
print(len(row))
j = 0
bus = Business.objects.get(id=arrID2bid[j])
for cell in row:
usr = User.objects.get(id=arrID2uid[i])
p = Recommendation(business=bus,recommendation=cell,username=usr)
predictions.append(p)
j+=1
i+=1
Recommendation.objects.bulk_create(predictions)
transaction.commit();
