def kFoldEvaluation(k, sessFile, featFile, weightFile, percent, typeFile):
sessions = loadSessions(sessFile)
#weightMatrix = readWeightMatrix(weightFile)
#
#p1 = {}
#r1 = {}
#p2 = {}
#r2 = {}
#p3 = {}
#r3 = {}
#p4 = {}
#r4 = {}
#
amean = []
ymean = []
for i in range(k):
x, y, uniqx, uniqy = sampleSessions(sessions, percent)
acount = 0.0
ylen = 0.0
termList, termDict = getTermList(uniqx)
for session in y:
aTerms, rTerms = addedAndRemovedTerms(session[0], session[1:], termDict)
acount += len(aTerms)
ylen += len(session)
print acount, ylen, acount / len(y), ylen / len(y)
amean.append(acount / len(y))
ymean.append(ylen / len(y))
print np.mean(amean), np.mean(ymean)
def predictTerms(queryList, y, qclusters):
termList, termDict = getTermList(queryList)
oracle_prec = 0.0
oracle_mrr = 0.0
added = 0
cScorer = ScoreClusterTerms()
for session in y:
query = session[0]
aTerms, rTerms = addedAndRemovedTerms(query, session[1:], termDict)
if len(aTerms) > 0:
prec1, mrr1 = getPrecRecall(termList, aTerms)
added += 1.0
oracle_prec += prec1
oracle_mrr += mrr1
print 'Oracle prec and recall ', oracle_prec / added, oracle_mrr / added, added
#porter = stem.porter.PorterStemmer();
clusters, clusIndex = toTerms(qclusters)
lim = 5
i = 0
prec = {}
mrr = {}
pf = 0.0
pr = 0.0
for session in y:
query = session[0].strip()
qSet = getQueryTerms(query) #getQueryTermsStemmed(query, porter);
aTerms, rTerms = addedAndRemovedTerms(query, session[1:], termDict)
if len(aTerms) > 0:
terms = cScorer.scoreWithCosine(qSet, clusters, clusIndex, lim)
if len(terms) > 0:
#print len(aTerms), len(terms)
prec1, mrr1 = getClustPrecRecall(terms, aTerms) # returns a list
#print 'METRIC',i, prec1, mrr1
#print topk , prec1, mrr1
if sum(prec1) > 0:
pf += 1.0
if sum(mrr1) > 0:
pr += 1.0
for topk in range(len(prec1)):
if topk not in prec:
prec[topk] = []
mrr[topk] = []
prec[topk].append(prec1[topk])
mrr[topk].append(mrr1[topk])
i += 1
retPrec = {}
retRecall = {}
for entry, ls in prec.items():
print 'Prec @', entry, np.mean(ls)
retPrec[entry] = np.mean(ls)
for entry, ls in mrr.items():
print 'Recall @', entry, np.mean(ls)
retRecall[entry] = np.mean(ls)
print 'Percentage ', pf / i, pr / i
return retPrec, retRecall
