def quantify(X_train, X_test, y_train, y_test, method):
q = Quantification(method=method)
q.fit(X_train, y_train)
KLD = q.score([X_test], [y_test])
#print('Test',q._classify_and_count([y_test]))
#print('Train',q._classify_and_count([y_train]))
return KLD
def quantify(X_train, X_test, y_train, y_test,method):
q=Quantification(method=method)
q.fit(X_train,y_train)
KLD=q.score([X_test], [y_test])
#print('Test',q._classify_and_count([y_test]))
#print('Train',q._classify_and_count([y_train]))
return KLD
def quantify_OHSUMED():
q = Quantification(method='test',
dir_name='QuantOHSUMED') #QuantRCV1 #QuantOHSUMED
[y_train, y_test_list, pred_prob_list, test_files,
y_names] = q._read_pickle('texts/cl_prob_' + q.dir_name + '.pickle')
indexes = q._read_pickle('texts/cl_indexes_' + q.dir_name + '.pickle')
td = q._classify_and_count(y_test_list)
#ed=q._expectation_maximization(y_train, pred_prob_list, stop_delta=0.1)
ed = q._exp_max(y_train, pred_prob_list, stop_delta=0.1)
#ed=q._prob_classify_and_count(pred_prob_list)
r = q._count_diff(td, ed)
return r
def quantify_OHSUMED():
q=Quantification(method='test', dir_name='QuantOHSUMED') #QuantRCV1 #QuantOHSUMED
[y_train,y_test_list,pred_prob_list,test_files, y_names]=q._read_pickle('texts/cl_prob_'+q.dir_name+'.pickle')
indexes=q._read_pickle('texts/cl_indexes_'+q.dir_name+'.pickle')
td=q._classify_and_count(y_test_list)
#ed=q._expectation_maximization(y_train, pred_prob_list, stop_delta=0.1)
ed=q._exp_max(y_train, pred_prob_list, stop_delta=0.1)
#ed=q._prob_classify_and_count(pred_prob_list)
r=q._count_diff(td, ed)
return r
def test_gen_data():
k_CC=0
k_PCC=0
k_ACC=0
k_PACC=0
k_EM=0
k_SVMp=0
k_Iter=0
k_EM1=0
CC=[]
PCC=[]
ACC=[]
PACC=[]
EM=[]
EM1=[]
SVMp=[]
Iter=[]
Iter1=[]
for i in range(1):
X,y=generate_data(n_samples=10000)
X_train, X_test, y_train, y_test=train_test_split(X,y, test_size=0.75)
q=Quantification(is_clean=True)
X_test_d,y_test_d=Quantification.make_drift_list(X_test,y_test,0.75)
q.fit(X_train,y_train)
#s = SVMperf(X_train, y_train, X_test_d, y_test_d)
#SVMp.append(q._kld(q._classify_and_count([y_test]), q._classify_and_count([s.getPredictions()])))
CC.append(q.score(X_test_d, y_test_d, method='CC'))
PCC.append(q.score(X_test_d, y_test_d, method='PCC'))
ACC.append(q.score(X_test_d, y_test_d, method='ACC'))
PACC.append(q.score(X_test_d, y_test_d, method='PACC'))
EM1.append(q.score(X_test_d, y_test_d, method='EM1'))
EM.append(q.score(X_test_d, y_test_d, method='EM'))
Iter.append(q.score(X_test_d, y_test_d, method='Iter'))
Iter1.append(q.score(X_test_d, y_test_d, method='Iter1'))
#print(i, q._classify_and_count([y_test]))
if Iter1[i]<=PCC[i]: k_PCC+=1
if Iter1[i]<=CC[i]: k_CC+=1
if Iter1[i]<=ACC[i]: k_ACC+=1
if Iter1[i]<=PACC[i]: k_PACC+=1
if Iter1[i]<=EM[i]: k_EM+=1
#if EM1[i]<=SVMp[i]: k_SVMp+=1
if Iter1[i]<=EM1[i]: k_EM1+=1
if Iter1[i]<=Iter[i]: k_Iter+=1
#print(q.predict(X_test_d, method='CC'),'\n',q.predict(X_test_d, method='PCC'),'\n',q.predict(X_test_d, method='ACC'),'\n',q.predict(X_test_d, method='PACC'),'\n',q.predict(X_test_d, method='EM1'),'\n',q.predict(X_test_d, method='EM'),'\n',SVMp,'\n',q.predict(X_test_d, method='Iter'))
print(i, q._classify_and_count(y_test_d),'\n')
print('CC\t','PCC\t','ACC\t','PACC\t','EM1\t','EM\t','SVMperf\t','Iter\t','Iter1\t')
print(np.average(CC),'\t',np.average(PCC),'\t',np.average(ACC),'\t',np.average(PACC),'\t',np.average(EM1),'\t',np.average(EM),'\t',np.average(SVMp),'\t',np.average(Iter),'\t',np.average(Iter1),'\t\t\t',
np.var(CC),'\t',np.var(PCC),'\t',np.var(ACC),'\t',np.var(PACC),'\t',np.var(EM1),'\t',np.var(EM),'\t',np.var(SVMp),'\t',np.var(Iter),'\t',np.var(Iter1),'\t\t\t',
k_CC,'\t',k_PCC,'\t',k_ACC,'\t',k_PACC,'\t',k_EM1,'\t',k_EM,'\t',k_SVMp,'\t',k_Iter)
def after_semEval():
#fname='100_topics_100_tweets.sentence-three-point.subtask-A'
#fname='100_topics_100_tweets.topic-five-point.subtask-CE'
fname='100_topics_XXX_tweets.topic-two-point.subtask-BD'
train=read_semeval('texts/2download/gold/all_train/'+fname+'.all.gold.tsv')
#train=read_semeval('texts/2download/gold/all/'+fname+'.train.gold.tsv')
tp=text_processing(n=1)
#cp=char_processing()
Xw_train=tp.fit_transform(train[0])
#pos_score_train, neg_score_train=readSentiScores('texts/2download/train+.out')
#add1=csr_matrix(pos_score_train).transpose()
#add2=csr_matrix(neg_score_train).transpose()
#Xw_train=scipy.sparse.hstack((scipy.sparse.hstack((Xw_train,add1)),add2), format='csr')
#Xc_train=cp.fit_transform(train[0])
#X_train=scipy.sparse.hstack((Xw_train, Xc_train), format='csr')
X_train=Xw_train
y_train=np.asarray(train[1])
#test=read_semeval('texts/2download/test_datasets-v2.0/SemEval2016-task4-test.subtask-BD.txt')
test=read_semeval('texts/2download/gold/test/test_gold_2.csv')
#test=read_semeval('texts/2download/gold/dev/'+fname+'.dev.gold.tsv')
Xw_test=tp.transform(test[0])
#pos_score_test, neg_score_test=readSentiScores('texts/2download/test+.out')
#add1=csr_matrix(pos_score_test).transpose()
#add2=csr_matrix(neg_score_test).transpose()
#Xw_test=scipy.sparse.hstack((scipy.sparse.hstack((Xw_test,add1)),add2), format='csr')
#Xc_test=cp.transform(test[0])
#X_test=scipy.sparse.hstack((Xw_test, Xc_test), format='csr')
X_test=Xw_test
y_test=np.asarray(test[1])
#perf=SVMperf(x_train=X_train, y_train=y_train, x_test=X_test, y_test=y_test)
#X_train, X_test, y_train, y_test=train_test_split(X_train,y_train, test_size=0.75)
print(X_test.shape)
X_test_list, y_test_list, utopics=split_by_topic(X_test, y_test, test[2])
q=Quantification(method='Iter1',is_clean=True)
#X_test, y_test=Quantification.make_drift_list(X_test.toarray(), y_test, proportion=0.2)
q.fit(X_train, y_train)
print('train',q._classify_and_count(y_train))
#prevs=q.predict_set(X_test_list, method='Iter1')
#write_semeval(dict(zip(utopics,prevs)),fname=fname)
#forSVMperf(q,y_test,test[2],y_test_list, utopics)
#q.iter_model.predict()
print('CC',q.score(X_test_list,y_test_list, method='CC'))
print('PCC',q.score(X_test_list,y_test_list, method='PCC'))
print('EM',q.score(X_test_list,y_test_list, method='EM'))
print('EM1',q.score(X_test_list,y_test_list, method='EM1'))
print('Iter',q.score(X_test_list,y_test_list, method='Iter'))
print('Iter1',q.score(X_test_list,y_test_list, method='Iter1'))
print('ACC',q.score(X_test_list,y_test_list, method='ACC'))
print('PACC',q.score(X_test_list,y_test_list, method='PACC'))
def test_gen_data():
k_CC = 0
k_PCC = 0
k_ACC = 0
k_PACC = 0
k_EM = 0
k_SVMp = 0
k_Iter = 0
k_EM1 = 0
CC = []
PCC = []
ACC = []
PACC = []
EM = []
EM1 = []
SVMp = []
Iter = []
Iter1 = []
for i in range(1):
X, y = generate_data(n_samples=10000)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.75)
q = Quantification(is_clean=True)
X_test_d, y_test_d = Quantification.make_drift_list(
X_test, y_test, 0.75)
q.fit(X_train, y_train)
#s = SVMperf(X_train, y_train, X_test_d, y_test_d)
#SVMp.append(q._kld(q._classify_and_count([y_test]), q._classify_and_count([s.getPredictions()])))
CC.append(q.score(X_test_d, y_test_d, method='CC'))
PCC.append(q.score(X_test_d, y_test_d, method='PCC'))
ACC.append(q.score(X_test_d, y_test_d, method='ACC'))
PACC.append(q.score(X_test_d, y_test_d, method='PACC'))
EM1.append(q.score(X_test_d, y_test_d, method='EM1'))
EM.append(q.score(X_test_d, y_test_d, method='EM'))
Iter.append(q.score(X_test_d, y_test_d, method='Iter'))
Iter1.append(q.score(X_test_d, y_test_d, method='Iter1'))
#print(i, q._classify_and_count([y_test]))
if Iter1[i] <= PCC[i]: k_PCC += 1
if Iter1[i] <= CC[i]: k_CC += 1
if Iter1[i] <= ACC[i]: k_ACC += 1
if Iter1[i] <= PACC[i]: k_PACC += 1
if Iter1[i] <= EM[i]: k_EM += 1
#if EM1[i]<=SVMp[i]: k_SVMp+=1
if Iter1[i] <= EM1[i]: k_EM1 += 1
if Iter1[i] <= Iter[i]: k_Iter += 1
#print(q.predict(X_test_d, method='CC'),'\n',q.predict(X_test_d, method='PCC'),'\n',q.predict(X_test_d, method='ACC'),'\n',q.predict(X_test_d, method='PACC'),'\n',q.predict(X_test_d, method='EM1'),'\n',q.predict(X_test_d, method='EM'),'\n',SVMp,'\n',q.predict(X_test_d, method='Iter'))
print(i, q._classify_and_count(y_test_d), '\n')
print('CC\t', 'PCC\t', 'ACC\t', 'PACC\t', 'EM1\t', 'EM\t', 'SVMperf\t',
'Iter\t', 'Iter1\t')
print(np.average(CC), '\t', np.average(PCC), '\t', np.average(ACC), '\t',
np.average(PACC), '\t', np.average(EM1), '\t', np.average(EM), '\t',
np.average(SVMp), '\t', np.average(Iter), '\t', np.average(Iter1),
'\t\t\t', np.var(CC), '\t', np.var(PCC), '\t', np.var(ACC), '\t',
np.var(PACC), '\t', np.var(EM1), '\t',
np.var(EM), '\t', np.var(SVMp), '\t', np.var(Iter), '\t',
np.var(Iter1), '\t\t\t', k_CC, '\t', k_PCC, '\t', k_ACC, '\t',
k_PACC, '\t', k_EM1, '\t', k_EM, '\t', k_SVMp, '\t', k_Iter)
def after_semEval():
#fname='100_topics_100_tweets.sentence-three-point.subtask-A'
#fname='100_topics_100_tweets.topic-five-point.subtask-CE'
fname = '100_topics_XXX_tweets.topic-two-point.subtask-BD'
train = read_semeval('texts/2download/gold/all_train/' + fname +
'.all.gold.tsv')
#train=read_semeval('texts/2download/gold/all/'+fname+'.train.gold.tsv')
tp = text_processing(n=1)
#cp=char_processing()
Xw_train = tp.fit_transform(train[0])
#pos_score_train, neg_score_train=readSentiScores('texts/2download/train+.out')
#add1=csr_matrix(pos_score_train).transpose()
#add2=csr_matrix(neg_score_train).transpose()
#Xw_train=scipy.sparse.hstack((scipy.sparse.hstack((Xw_train,add1)),add2), format='csr')
#Xc_train=cp.fit_transform(train[0])
#X_train=scipy.sparse.hstack((Xw_train, Xc_train), format='csr')
X_train = Xw_train
y_train = np.asarray(train[1])
#test=read_semeval('texts/2download/test_datasets-v2.0/SemEval2016-task4-test.subtask-BD.txt')
test = read_semeval('texts/2download/gold/test/test_gold_2.csv')
#test=read_semeval('texts/2download/gold/dev/'+fname+'.dev.gold.tsv')
Xw_test = tp.transform(test[0])
#pos_score_test, neg_score_test=readSentiScores('texts/2download/test+.out')
#add1=csr_matrix(pos_score_test).transpose()
#add2=csr_matrix(neg_score_test).transpose()
#Xw_test=scipy.sparse.hstack((scipy.sparse.hstack((Xw_test,add1)),add2), format='csr')
#Xc_test=cp.transform(test[0])
#X_test=scipy.sparse.hstack((Xw_test, Xc_test), format='csr')
X_test = Xw_test
y_test = np.asarray(test[1])
#perf=SVMperf(x_train=X_train, y_train=y_train, x_test=X_test, y_test=y_test)
#X_train, X_test, y_train, y_test=train_test_split(X_train,y_train, test_size=0.75)
print(X_test.shape)
X_test_list, y_test_list, utopics = split_by_topic(X_test, y_test, test[2])
q = Quantification(method='Iter1', is_clean=True)
#X_test, y_test=Quantification.make_drift_list(X_test.toarray(), y_test, proportion=0.2)
q.fit(X_train, y_train)
print('train', q._classify_and_count(y_train))
#prevs=q.predict_set(X_test_list, method='Iter1')
#write_semeval(dict(zip(utopics,prevs)),fname=fname)
#forSVMperf(q,y_test,test[2],y_test_list, utopics)
#q.iter_model.predict()
print('CC', q.score(X_test_list, y_test_list, method='CC'))
print('PCC', q.score(X_test_list, y_test_list, method='PCC'))
print('EM', q.score(X_test_list, y_test_list, method='EM'))
print('EM1', q.score(X_test_list, y_test_list, method='EM1'))
print('Iter', q.score(X_test_list, y_test_list, method='Iter'))
print('Iter1', q.score(X_test_list, y_test_list, method='Iter1'))
print('ACC', q.score(X_test_list, y_test_list, method='ACC'))
print('PACC', q.score(X_test_list, y_test_list, method='PACC'))
