def bootstrap():
#dataset = Processing().import_data()
for dataset, filename in Processing().import_single_data():
print(filename)
training_data_X, training_data_y, testing_data_X, testing_data_y = Processing(
).separate_data(dataset)
# print('train shape', training_data_X.shape)
training_data_X = training_data_X.tolist()
training_data_y = training_data_y.tolist()
from PyOptimize.General_Opt import Test_function
def LTR(a, **kwargs):
return Test_function().LTR(a, **kwargs)
ga = pyGaft(objfunc=LTR,
var_bounds=[(-2, 2)] * 20,
individual_size=50,
max_iter=10,
max_or_min='max',
X=training_data_X,
y=training_data_y).run()
importlib.reload(best_fit)
a, fitness_value = best_fit.best_fit[-1][-2], best_fit.best_fit[-1][-1]
print('a = {0}'.format(a))
pred_y = []
for test_x in testing_data_X:
pred_y.append(np.dot(test_x, a))
fpa = PerformanceMeasure(testing_data_y, pred_y).FPA()
print('fpa = {0}'.format(fpa))
def bootstrap(dataset):
count = 0
training_data_X, training_data_y, testing_data_X, testing_data_y = Processing(
).separate_data(dataset)
# cost sensitive ranking SVM
csrs = RankSVM()
P, r = csrs.transform_pairwise(training_data_X, training_data_y)
P = P.tolist()
r = r.tolist()
u, n = PerformanceMeasure(training_data_y).calc_UN(type='cs')
count += 1
global Loss
csga = pyGaft(objfunc=Loss,
var_bounds=[(-1, 1)] * 20,
individual_size=500,
max_iter=2,
max_or_min='min',
P=P,
r=r,
u=u,
n=n).run()
if count == 1:
import best_fit
else:
importlib.reload(best_fit)
w, fitness = best_fit.best_fit[-1][-2], best_fit.best_fit[-1][-1]
#print('w = ', w)
csrs_pred_y = RankSVM(w=w).predict3(testing_data_X)
csrs_fpa = PerformanceMeasure(testing_data_y, csrs_pred_y).FPA()
csrs_fpa_list.append(csrs_fpa)
csrs_pofb = PerformanceMeasure(testing_data_y, csrs_pred_y).PofB20()
csrs_pofd = PerformanceMeasure(testing_data_y, csrs_pred_y).PofD20()
csrs_ranking = PerformanceMeasure(testing_data_y, csrs_pred_y).ranking()
csrs_pofb_list.append(csrs_pofb)
csrs_pofd_list.append(csrs_pofd)
csrs_ranking_list.append(csrs_ranking)
# IR SVM
u, n = PerformanceMeasure(training_data_y).calc_UN(type='ir')
count += 1
irga = pyGaft(objfunc=Loss,
var_bounds=[(-1, 1)] * 20,
individual_size=500,
max_iter=2,
max_or_min='min',
P=P,
r=r,
u=u,
n=n).run()
if count == 1:
import best_fit
else:
importlib.reload(best_fit)
w, fitness = best_fit.best_fit[-1][-2], best_fit.best_fit[-1][-1]
#print('w = ', w)
irsvm_pred_y = RankSVM(w=w).predict3(testing_data_X)
irsvm_fpa = PerformanceMeasure(testing_data_y, irsvm_pred_y).FPA()
#print('irsvm_fpa:', irsvm_fpa)
irsvm_fpa_list.append(irsvm_fpa)
irsvm_pofb = PerformanceMeasure(testing_data_y, irsvm_pred_y).PofB20()
irsvm_pofd = PerformanceMeasure(testing_data_y, irsvm_pred_y).PofD20()
irsvm_ranking = PerformanceMeasure(testing_data_y, irsvm_pred_y).ranking()
irsvm_pofb_list.append(irsvm_pofb)
irsvm_pofd_list.append(irsvm_pofd)
irsvm_ranking_list.append(irsvm_ranking)
# 这里还要加个去掉另一个参数的
u, n = PerformanceMeasure(training_data_y).calc_UN(type='svm')
count += 1
irga = pyGaft(objfunc=Loss,
var_bounds=[(-1, 1)] * 20,
individual_size=500,
max_iter=2,
max_or_min='min',
P=P,
r=r,
u=u,
n=n).run()
if count == 1:
import best_fit
else:
importlib.reload(best_fit)
w, fitness = best_fit.best_fit[-1][-2], best_fit.best_fit[-1][-1]
#print('w = ', w)
svm_pred_y = RankSVM(w=w).predict3(testing_data_X)
svm_fpa = PerformanceMeasure(testing_data_y, svm_pred_y).FPA()
#print('svm_fpa:', svm_fpa)
svm_fpa_list.append(svm_fpa)
svm_pofb = PerformanceMeasure(testing_data_y, svm_pred_y).PofB20()
svm_pofd = PerformanceMeasure(testing_data_y, svm_pred_y).PofD20()
svm_ranking = PerformanceMeasure(testing_data_y, svm_pred_y).ranking()
svm_pofb_list.append(svm_pofb)
svm_pofd_list.append(svm_pofd)
svm_ranking_list.append(svm_ranking)
# 这个是LTR
training_datalist_X = training_data_X.tolist()
training_datalist_y = training_data_y.tolist()
from PyOptimize.General_Opt import Test_function
count += 1
global LTR
ltrga = pyGaft(objfunc=LTR,
var_bounds=[(-20, 20)] * 20,
individual_size=100,
max_iter=2,
max_or_min='max',
X=training_datalist_X,
y=training_datalist_y).run()
if count == 1:
import best_fit
else:
importlib.reload(best_fit)
a, fitness_value = best_fit.best_fit[-1][-2], best_fit.best_fit[-1][-1]
#print('a = {0}'.format(a))
ltr_pred_y = []
for test_x in testing_data_X:
ltr_pred_y.append(np.dot(test_x, a))
ltr_fpa = PerformanceMeasure(testing_data_y, ltr_pred_y).FPA()
# print('ltr_fpa', ltr_fpa)
ltr_fpa_list.append(ltr_fpa)
ltr_pofb = PerformanceMeasure(testing_data_y, ltr_pred_y).PofB20()
ltr_pofd = PerformanceMeasure(testing_data_y, ltr_pred_y).PofD20()
ltr_ranking = PerformanceMeasure(testing_data_y, ltr_pred_y).ranking()
ltr_pofb_list.append(ltr_pofb)
ltr_pofd_list.append(ltr_pofd)
ltr_ranking_list.append(ltr_ranking)
# 在原始数据集上训练Ranking SVM,DTR,LR,BRR
# 这里加上了shuffle，是为了让r的值不全为1，全为1svm会报错
shuf_X, shuf_y = shuffle(training_data_X, training_data_y)
rs = RankSVM(C=1.0).fit(shuf_X, shuf_y)
rs_pred_y = np.around(rs.predict2(testing_data_X))
rs_fpa = PerformanceMeasure(testing_data_y, rs_pred_y).FPA()
# print('rs_fpa:', rs_fpa)
rs_fpa_list.append(rs_fpa)
rs_pofb = PerformanceMeasure(testing_data_y, rs_pred_y).PofB20()
rs_pofd = PerformanceMeasure(testing_data_y, rs_pred_y).PofD20()
rs_ranking = PerformanceMeasure(testing_data_y, rs_pred_y).ranking()
rs_pofb_list.append(rs_pofb)
rs_pofd_list.append(rs_pofd)
rs_ranking_list.append(rs_ranking)
dtr = DecisionTreeRegressor().fit(training_data_X, training_data_y)
dtr_pred_y = dtr.predict(testing_data_X)
dtr_fpa = PerformanceMeasure(testing_data_y, dtr_pred_y).FPA()
# print('dtr_fpa:', dtr_fpa)
dtr_fpa_list.append(dtr_fpa)
dtr_pofb = PerformanceMeasure(testing_data_y, dtr_pred_y).PofB20()
dtr_pofd = PerformanceMeasure(testing_data_y, dtr_pred_y).PofD20()
dtr_ranking = PerformanceMeasure(testing_data_y, dtr_pred_y).ranking()
dtr_pofb_list.append(dtr_pofb)
dtr_pofd_list.append(dtr_pofd)
dtr_ranking_list.append(dtr_ranking)
lr = linear_model.LinearRegression().fit(training_data_X, training_data_y)
lr_pred_y = lr.predict(testing_data_X)
lr_fpa = PerformanceMeasure(testing_data_y, lr_pred_y).FPA()
# print('lr_fpa:', lr_fpa)
lr_fpa_list.append(lr_fpa)
lr_pofb = PerformanceMeasure(testing_data_y, lr_pred_y).PofB20()
lr_pofd = PerformanceMeasure(testing_data_y, lr_pred_y).PofD20()
lr_ranking = PerformanceMeasure(testing_data_y, lr_pred_y).ranking()
lr_pofb_list.append(lr_pofb)
lr_pofd_list.append(lr_pofd)
lr_ranking_list.append(lr_ranking)
brr = BayesianRidge().fit(training_data_X, training_data_y)
brr_pred_y = brr.predict(testing_data_X)
brr_fpa = PerformanceMeasure(testing_data_y, brr_pred_y).FPA()
# print('brr_fpa:', brr_fpa)
brr_fpa_list.append(brr_fpa)
brr_pofb = PerformanceMeasure(testing_data_y, brr_pred_y).PofB20()
brr_pofd = PerformanceMeasure(testing_data_y, brr_pred_y).PofD20()
brr_ranking = PerformanceMeasure(testing_data_y, brr_pred_y).ranking()
brr_pofb_list.append(brr_pofb)
brr_pofd_list.append(brr_pofd)
brr_ranking_list.append(brr_ranking)
# 先对训练数据集进行RUS处理，然后训练Ranking SVM, DTR,LR,BRR
rus_X, rus_y, _id = RandomUnderSampler(ratio=1.0,
return_indices=True).fit_sample(
training_data_X,
training_data_y)
# LTR
training_datalist_X = rus_X.tolist()
training_datalist_y = rus_y.tolist()
from PyOptimize.General_Opt import Test_function
count += 1
rus_ltrga = pyGaft(objfunc=LTR,
var_bounds=[(-20, 20)] * 20,
individual_size=100,
max_iter=2,
max_or_min='max',
X=training_datalist_X,
y=training_datalist_y).run()
if count == 1:
import best_fit
else:
importlib.reload(best_fit)
rus_a, fitness_value = best_fit.best_fit[-1][-2], best_fit.best_fit[-1][-1]
#print('rus_a = {0}'.format(rus_a))
rus_ltr_pred_y = []
for test_x in testing_data_X:
rus_ltr_pred_y.append(np.dot(test_x, rus_a))
rus_ltr_fpa = PerformanceMeasure(testing_data_y, rus_ltr_pred_y).FPA()
# print('rus_ltr_fpa', rus_ltr_fpa)
rus_ltr_fpa_list.append(rus_ltr_fpa)
rus_ltr_pofb = PerformanceMeasure(testing_data_y, rus_ltr_pred_y).PofB20()
rus_ltr_pofd = PerformanceMeasure(testing_data_y, rus_ltr_pred_y).PofD20()
rus_ltr_ranking = PerformanceMeasure(testing_data_y,
rus_ltr_pred_y).ranking()
rus_ltr_pofb_list.append(rus_ltr_pofb)
rus_ltr_pofd_list.append(rus_ltr_pofd)
rus_ltr_ranking_list.append(rus_ltr_ranking)
shuf_X, shuf_y = shuffle(rus_X, rus_y)
rus_rs = RankSVM(C=1.0).fit(shuf_X, shuf_y)
rus_rs_pred_y = rus_rs.predict2(testing_data_X)
rus_rs_fpa = PerformanceMeasure(testing_data_y, rus_rs_pred_y).FPA()
# print('rus_rs_fpa:', rus_rs_fpa)
rus_rs_fpa_list.append(rus_rs_fpa)
rus_rs_pofb = PerformanceMeasure(testing_data_y, rus_rs_pred_y).PofB20()
rus_rs_pofd = PerformanceMeasure(testing_data_y, rus_rs_pred_y).PofD20()
rus_rs_ranking = PerformanceMeasure(testing_data_y,
rus_rs_pred_y).ranking()
rus_rs_pofb_list.append(rus_rs_pofb)
rus_rs_pofd_list.append(rus_rs_pofd)
rus_rs_ranking_list.append(rus_rs_ranking)
rus_dtr = DecisionTreeRegressor().fit(rus_X, rus_y)
rus_dtr_pred_y = rus_dtr.predict(testing_data_X)
rus_dtr_fpa = PerformanceMeasure(testing_data_y, rus_dtr_pred_y).FPA()
# print('rus_dtr_fpa:', rus_dtr_fpa)
rus_dtr_fpa_list.append(rus_dtr_fpa)
rus_dtr_pofb = PerformanceMeasure(testing_data_y, rus_dtr_pred_y).PofB20()
rus_dtr_pofd = PerformanceMeasure(testing_data_y, rus_dtr_pred_y).PofD20()
rus_dtr_ranking = PerformanceMeasure(testing_data_y,
rus_dtr_pred_y).ranking()
rus_dtr_pofb_list.append(rus_dtr_pofb)
rus_dtr_pofd_list.append(rus_dtr_pofd)
rus_dtr_ranking_list.append(rus_dtr_ranking)
rus_lr = linear_model.LinearRegression().fit(rus_X, rus_y)
rus_lr_pred_y = rus_lr.predict(testing_data_X)
rus_lr_fpa = PerformanceMeasure(testing_data_y, rus_lr_pred_y).FPA()
# print('rus_lr_fpa:', rus_lr_fpa)
rus_lr_fpa_list.append(rus_lr_fpa)
rus_lr_pofb = PerformanceMeasure(testing_data_y, rus_lr_pred_y).PofB20()
rus_lr_pofd = PerformanceMeasure(testing_data_y, rus_lr_pred_y).PofD20()
rus_lr_ranking = PerformanceMeasure(testing_data_y,
rus_lr_pred_y).ranking()
rus_lr_pofb_list.append(rus_lr_pofb)
rus_lr_pofd_list.append(rus_lr_pofd)
rus_lr_ranking_list.append(rus_lr_ranking)
rus_brr = BayesianRidge().fit(rus_X, rus_y)
rus_brr_pred_y = rus_brr.predict(testing_data_X)
rus_brr_fpa = PerformanceMeasure(testing_data_y, rus_brr_pred_y).FPA()
# print('rus_brr_fpa:', rus_brr_fpa)
rus_brr_fpa_list.append(rus_brr_fpa)
rus_brr_pofb = PerformanceMeasure(testing_data_y, rus_brr_pred_y).PofB20()
rus_brr_pofd = PerformanceMeasure(testing_data_y, rus_brr_pred_y).PofD20()
rus_brr_ranking = PerformanceMeasure(testing_data_y,
rus_brr_pred_y).ranking()
rus_brr_pofb_list.append(rus_brr_pofb)
rus_brr_pofd_list.append(rus_brr_pofd)
rus_brr_ranking_list.append(rus_brr_ranking)
# 先对训练数据集进行Smote处理，然后训练Ranking SVM, DTR,LR,BRR
smote_X, smote_y = Smote(training_data_X, training_data_y, ratio=1.0,
k=5).over_sampling_addorginaldata()
# LTR
training_datalist_X = smote_X.tolist()
training_datalist_y = smote_y.tolist()
from PyOptimize.General_Opt import Test_function
count += 1
smote_ltrga = pyGaft(objfunc=LTR,
var_bounds=[(-20, 20)] * 20,
individual_size=100,
max_iter=2,
max_or_min='max',
X=training_datalist_X,
y=training_datalist_y).run()
if count == 1:
import best_fit
else:
importlib.reload(best_fit)
smote_a, fitness_value = best_fit.best_fit[-1][-2], best_fit.best_fit[-1][
-1]
smote_ltr_pred_y = []
for test_x in testing_data_X:
smote_ltr_pred_y.append(np.dot(test_x, smote_a))
smote_ltr_fpa = PerformanceMeasure(testing_data_y, smote_ltr_pred_y).FPA()
smote_ltr_fpa_list.append(smote_ltr_fpa)
smote_ltr_pofb = PerformanceMeasure(testing_data_y,
smote_ltr_pred_y).PofB20()
smote_ltr_pofd = PerformanceMeasure(testing_data_y,
smote_ltr_pred_y).PofD20()
smote_ltr_ranking = PerformanceMeasure(testing_data_y,
smote_ltr_pred_y).ranking()
smote_ltr_pofb_list.append(smote_ltr_pofb)
smote_ltr_pofd_list.append(smote_ltr_pofd)
smote_ltr_ranking_list.append(smote_ltr_ranking)
shuf_X, shuf_y = shuffle(smote_X, smote_y)
smote_rs = RankSVM(C=1.0).fit(shuf_X, shuf_y)
smote_rs_pred_y = smote_rs.predict2(testing_data_X)
smote_rs_fpa = PerformanceMeasure(testing_data_y, smote_rs_pred_y).FPA()
smote_rs_fpa_list.append(smote_rs_fpa)
smote_rs_pofb = PerformanceMeasure(testing_data_y,
smote_rs_pred_y).PofB20()
smote_rs_pofd = PerformanceMeasure(testing_data_y,
smote_rs_pred_y).PofD20()
smote_rs_ranking = PerformanceMeasure(testing_data_y,
smote_rs_pred_y).ranking()
smote_rs_pofb_list.append(smote_rs_pofb)
smote_rs_pofd_list.append(smote_rs_pofd)
smote_rs_ranking_list.append(smote_rs_ranking)
smote_dtr = DecisionTreeRegressor().fit(smote_X, smote_y)
smote_dtr_pred_y = smote_dtr.predict(testing_data_X)
smote_dtr_fpa = PerformanceMeasure(testing_data_y, smote_dtr_pred_y).FPA()
smote_dtr_fpa_list.append(smote_dtr_fpa)
smote_dtr_pofb = PerformanceMeasure(testing_data_y,
smote_dtr_pred_y).PofB20()
smote_dtr_pofd = PerformanceMeasure(testing_data_y,
smote_dtr_pred_y).PofD20()
smote_dtr_ranking = PerformanceMeasure(testing_data_y,
smote_dtr_pred_y).ranking()
smote_dtr_pofb_list.append(smote_dtr_pofb)
smote_dtr_pofd_list.append(smote_dtr_pofd)
smote_dtr_ranking_list.append(smote_dtr_ranking)
smote_lr = linear_model.LinearRegression().fit(smote_X, smote_y)
smote_lr_pred_y = smote_lr.predict(testing_data_X)
smote_lr_fpa = PerformanceMeasure(testing_data_y, smote_lr_pred_y).FPA()
smote_lr_fpa_list.append(smote_lr_fpa)
smote_lr_pofb = PerformanceMeasure(testing_data_y,
smote_lr_pred_y).PofB20()
smote_lr_pofd = PerformanceMeasure(testing_data_y,
smote_lr_pred_y).PofD20()
smote_lr_ranking = PerformanceMeasure(testing_data_y,
smote_lr_pred_y).ranking()
smote_lr_pofb_list.append(smote_lr_pofb)
smote_lr_pofd_list.append(smote_lr_pofd)
smote_lr_rankig_list.append(smote_lr_rankig)
smote_brr = BayesianRidge().fit(smote_X, smote_y)
smote_brr_pred_y = smote_brr.predict(testing_data_X)
smote_brr_fpa = PerformanceMeasure(testing_data_y, smote_brr_pred_y).FPA()
smote_brr_fpa_list.append(smote_brr_fpa)
smote_brr_pofb = PerformanceMeasure(testing_data_y,
smote_brr_pred_y).PofB20()
smote_brr_pofd = PerformanceMeasure(testing_data_y,
smote_brr_pred_y).PofD20()
smote_brr_ranking = PerformanceMeasure(testing_data_y,
smote_brr_pred_y).ranking()
smote_brr_pofb_list.append(smote_brr_pofb)
smote_brr_pofd_list.append(smote_brr_pofd)
smote_brr_ranking_list.append(smote_brr_ranking)
