def nn_feature_selection_wrap(training_feature,training_label,alpha):
Result_List = CalParList (3,"alpha","feature_0","feature_1")
skf = StratifiedKFold(n_splits=10,shuffle=True)
skf.get_n_splits(training_feature,training_label['label'])
feature_avaliable = ['feature0','feature1','feature2','feature3','feature4']
feature_choice=list(itertools.combinations(feature_avaliable ,2))
for i in range(len(feature_choice)):
Cal_Result_List = CalList ()
for train_index, test_index in skf.split(training_feature,training_label['label']):
X_train, X_val = training_feature.loc[train_index], training_feature.loc[test_index]
y_train, y_val = training_label.loc[train_index], training_label.loc[test_index]
X_train=X_train.loc[:,[feature_choice[i][0],feature_choice[i][1]]]
X_val=X_val.loc[:,[feature_choice[i][0],feature_choice[i][1]]]
dis_1,dis_2=nn_distance_calculate(X_val,X_train,y_train)
y_pred_temp=nn_predict(dis_1,dis_2,alpha,X_val)
Precall,f1_score,BER,FPR = cal_score (y_pred_temp,y_val['label'])
Cal_Result_List.list_append(Precall,f1_score,BER,FPR)
Precall,FPR,BER,f1_score = Cal_Result_List.list_average_cal()
Result_List.list_append (Precall,f1_score,BER,FPR,alpha,feature_choice[i][0],feature_choice[i][1])
result = Result_List.return_result()
return result
def nn_validation(X_train, y_train, X_val, y_val, alpha):
dis_1, dis_2 = nn_distance_calculate(X_val, X_train, y_train)
y_pred_temp = nn_predict(dis_1, dis_2, alpha, X_val)
Precall, f1_score, BER, FPR = cal_score(y_pred_temp, y_val['label'])
print("TPR:" + str(Precall) + " f1 score:" + str(f1_score) + " FPR:" +
str(FPR) + " BER:" + str(BER))
return y_pred_temp
def kNN_k_parameter_adjust (X_train,y_train,X_val,y_val):
start=time.time()
k_value=5
K_list = []
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
time_list=[]
while (k_value<300):
start1=time.time()
neigh = KNeighborsClassifier(n_neighbors=k_value,algorithm='auto',weights ='distance')
neigh.fit(X_train, y_train['label'])
y_pred = neigh.predict(X_val)
y_pred = {"label_pred":y_pred}
y_pred = pd.DataFrame(data=y_pred)
Precall,f1_score,BER,FPR = cal_score (y_pred,y_val['label'])
K_list.append(k_value)
tpr_list.append(Precall)
fpr_list.append(FPR)
BER_list.append(BER)
f1_score_list.append(f1_score)
temp=(time.time()-start1)/60
time_list.append(temp)
#print("current k_value:%d ,fit time:%5.1fminute"%(k_value,(time.time()-start1)/60))
if (k_value<=20):
k_value=k_value+1
else:
k_value=k_value+3
#print(" ")
print("the total executing time:%5.1fminute"%((time.time()-start)/60))
result = {"k_value":K_list,"TPR":tpr_list,"FPR":fpr_list,"f1_score":f1_score_list,"BER":BER_list,"time":time_list}
columns = ["k_value","f1_score","TPR","FPR","BER","time"]
result = pd.DataFrame (data=result,columns=columns)
return result
def nn_predict_with_distance_adjust_presion(X_train, y_train, X_val, y_val,
alpha_lower_bound,
alpha_higher_bound):
dis_1, dis_2 = nn_distance_calculate(X_val, X_train, y_train)
alpha = alpha_lower_bound
alpha_list = []
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
while (alpha <= alpha_higher_bound):
#print ("current alpha:"+str(alpha))
y_pred_temp = nn_predict(dis_1, dis_2, alpha, X_val)
Precall, f1_score, BER, FPR = cal_score(y_pred_temp, y_val['label'])
alpha_list.append(alpha)
tpr_list.append(Precall)
fpr_list.append(FPR)
BER_list.append(BER)
f1_score_list.append(f1_score)
alpha = alpha + 0.001
result = {
"alpha": alpha_list,
"TPR": tpr_list,
"FPR": fpr_list,
"f1_score": f1_score_list,
"BER": BER_list
}
columns = ["alpha", "f1_score", "TPR", "FPR", "BER"]
result = pd.DataFrame(data=result, columns=columns)
return result
def best_result_choosen(dis_1, dis_2, X_val, y_val):
alpha = 0.47
alpha_list = []
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
while (alpha <= 0.57):
#print ("current alpha:"+str(alpha))
y_pred_temp = nn_predict(dis_1, dis_2, alpha, X_val)
Precall, f1_score, BER, FPR = cal_score(y_pred_temp, y_val['label'])
alpha_list.append(alpha)
tpr_list.append(Precall)
fpr_list.append(FPR)
BER_list.append(BER)
f1_score_list.append(f1_score)
alpha = alpha + 0.002
max_f1_score_index = f1_score_list.index(max(f1_score_list))
result = {
"alpha": alpha_list,
"TPR": tpr_list,
"FPR": fpr_list,
"f1_score": f1_score_list,
"BER": BER_list
}
columns = ["alpha", "f1_score", "TPR", "FPR", "BER"]
result = pd.DataFrame(data=result, columns=columns)
print(result.loc[result['f1_score'].idxmax()])
return 0
def SVM_base_fuction (X_train,y_train,X_val,y_val):
svc_clf=SVC(kernel="rbf",degree=len(X_train))
svc_clf.fit(X_train,y_train['label'])
y_pred=svc_clf.predict(X_val)
y_pred = {"label_pred":y_pred}
y_pred = pd.DataFrame(data=y_pred)
Precall,f1_score,BER,FPR= cal_score(y_pred,y_val['label'])
return Precall,f1_score,BER,FPR,y_pred
def kNN_base_function(X_train, y_train, X_val, y_val, k_value):
neigh = KNeighborsClassifier(n_neighbors=k_value,
algorithm='auto',
weights='distance')
neigh.fit(X_train, y_train['label'])
y_pred = neigh.predict(X_val)
y_pred = {"label_pred": y_pred}
y_pred = pd.DataFrame(data=y_pred)
Precall, f1_score, BER, FPR = cal_score(y_pred, y_val['label'])
return Precall, f1_score, BER, FPR, y_pred
def nn_predict_with_distance_adjust(training_feature,
training_label): #,#X_val,y_val):
alpha_list = []
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
skf = StratifiedKFold(n_splits=10, shuffle=True)
skf.get_n_splits(training_feature, training_label['label'])
alpha = 0.10
while (alpha <= 0.9):
tpr_list_temp = []
fpr_list_temp = []
BER_list_temp = []
f1_score_list_temp = []
for train_index, test_index in skf.split(training_feature,
training_label['label']):
X_train, X_val = training_feature.loc[
train_index], training_feature.loc[test_index]
y_train, y_val = training_label.loc[
train_index], training_label.loc[test_index]
dis_1, dis_2 = nn_distance_calculate(X_val, X_train, y_train)
#print ("current alpha:"+str(alpha))
y_pred_temp = nn_predict(dis_1, dis_2, alpha, X_val)
Precall, f1_score, BER, FPR = cal_score(y_pred_temp,
y_val['label'])
tpr_list_temp.append(Precall)
fpr_list_temp.append(FPR)
BER_list_temp.append(BER)
f1_score_list_temp.append(f1_score)
alpha_list.append(alpha)
tpr_list.append(sum(tpr_list_temp) / len(tpr_list_temp))
fpr_list.append(sum(fpr_list_temp) / len(fpr_list_temp))
BER_list.append(sum(BER_list_temp) / len(BER_list_temp))
f1_score_list.append(sum(f1_score_list_temp) / len(f1_score_list_temp))
if (0.4 <= alpha <= 0.6):
alpha = alpha + 0.01
else:
alpha = alpha + 0.1
result = {
"alpha": alpha_list,
"TPR": tpr_list,
"FPR": fpr_list,
"f1_score": f1_score_list,
"BER": BER_list
}
columns = ["alpha", "f1_score", "TPR", "FPR", "BER"]
result = pd.DataFrame(data=result, columns=columns)
return result
def kNN_k_parameter_adjust_with_bisaes_data (X_train,y_train,X_val,y_val,data_ratio):
start=time.time()
k_value=5
K_list = []
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
time_list=[]
train_data = {'feature0':X_train['feature0'],'feature1':X_train['feature1'],'feature2':X_train['feature2'],'feature3':X_train['feature3'],'feature4':X_train['feature4'],'label':y_train['label']}
train_data = pd.DataFrame(data=train_data)
Class1_sample =pd.DataFrame.sample(train_data[train_data['label']==1],int(8000*data_ratio))
Class2_sample =pd.DataFrame.sample(train_data[train_data['label']==2],8000)
res = [Class1_sample, Class2_sample]
train_com = pd.concat(res)
#print(train_com)
sample_label = pd.DataFrame(train_com['label'])
sample_feature=train_com.drop(["label"],axis=1)
while (k_value<300):
start1=time.time()
neigh = KNeighborsClassifier(n_neighbors=k_value,algorithm='auto',weights ='distance')
neigh.fit(sample_feature, sample_label['label'])
y_pred = neigh.predict(X_val)
y_pred = {"label_pred":y_pred}
y_pred = pd.DataFrame(data=y_pred)
Precall,f1_score,BER,FPR = cal_score (y_pred,y_val['label'])
K_list.append(k_value)
tpr_list.append(Precall)
fpr_list.append(FPR)
BER_list.append(BER)
f1_score_list.append(f1_score)
temp=(time.time()-start1)/60
time_list.append(temp)
#print("current k_value:%d ,fit time:%5.1fminute"%(k_value,(time.time()-start1)/60))
if (k_value<=20):
k_value=k_value+1
else:
k_value=k_value+3
#print(" ")
print("the total executing time:%5.1fminute"%((time.time()-start)/60))
result = {"k_value":K_list,"TPR":tpr_list,"FPR":fpr_list,"f1_score":f1_score_list,"BER":BER_list,"time":time_list}
columns = ["k_value","f1_score","TPR","FPR","BER","time"]
result = pd.DataFrame (data=result,columns=columns)
return result
def parameter_adjust_presion (X_train,y_train,X_val,y_val,sample_amount,data_ratio):
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
time_list=[]
gamma_exp_list=[]
C_exp_list=[]
label_1_amount = int(sample_amount * (data_ratio/(data_ratio+1)))
label_2_amount = int(sample_amount-label_1_amount)
C_exp = -5
while (C_exp <3):
gamma_exp = -C_exp-12
start1=time.time();
train_data = pd.concat([X_train,y_train['label']],axis=1,join='outer')
train_data = pd.DataFrame(train_data)
Class1_sample =pd.DataFrame.sample(train_data[train_data['label']==1],label_1_amount)
Class2_sample =pd.DataFrame.sample(train_data[train_data['label']==2],label_2_amount)
res = [Class1_sample, Class2_sample]
train_com = pd.concat(res)
sample_label = pd.DataFrame(train_com['label'])
sample_feature=train_com.drop(["label"],axis=1)
svc_clf=SVC(kernel="rbf",degree=len(sample_feature),gamma=2**gamma_exp,C=2**C_exp)
svc_clf.fit(sample_feature,sample_label['label'])
y_pred=svc_clf.predict(X_val)
y_pred = {"label_pred":y_pred}
y_pred = pd.DataFrame(data=y_pred)
#print ("curren gamma_exp:"+str(gamma_exp))
#print ("current C_exp:"+str(C_exp))
Precall,f1_score,BER,FPR= cal_score(y_pred,y_val['label'])
temp=(time.time()-start1)/60
time_list.append(temp)
tpr_list.append(Precall)
f1_score_list.append(f1_score)
BER_list.append(BER)
fpr_list.append(FPR)
gamma_exp_list.append(gamma_exp)
C_exp_list.append(C_exp)
#print("fit time:%5.1fminute"%(temp))
result = {"gamma_exp":gamma_exp_list,"C_exp":C_exp_list,"TPR":tpr_list,"FPR":fpr_list,"f1_score":f1_score_list,"BER":BER_list,"time":time_list}
columns = ["gamma_exp","C_exp","f1_score","TPR","FPR","BER","time"]
result = pd.DataFrame (data=result,columns=columns)
return result
def NN_cross_validation(training_feature,training_label,alpha):
skf = StratifiedKFold(n_splits=10,shuffle=True)
skf.get_n_splits(training_feature,training_label['label'])
Cal_Result_List = CalList ()
skf = StratifiedKFold(n_splits=10,shuffle=True)
skf.get_n_splits(training_feature,training_label['label'])
for train_index, test_index in skf.split(training_feature,training_label['label']):
X_train, X_val = training_feature.loc[train_index], training_feature.loc[test_index]
y_train, y_val = training_label.loc[train_index], training_label.loc[test_index]
dis_1,dis_2=nn_distance_calculate(X_val,X_train,y_train)
y_pred_temp=nn_predict(dis_1,dis_2,alpha,X_val)
Precall,f1_score,BER,FPR = cal_score (y_pred_temp,y_val['label'])
Cal_Result_List.list_append(Precall,f1_score,BER,FPR)
Precall,FPR,BER,f1_score = Cal_Result_List.list_average_cal()
return Precall,FPR,BER,f1_score
def nn_feature_selection_wrap(training_feature, training_label, alpha):
feature_0_list = []
feature_1_list = []
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
alpha_list = []
skf = StratifiedKFold(n_splits=10, shuffle=True)
skf.get_n_splits(training_feature, training_label['label'])
feature_avaliable = [
'feature0', 'feature1', 'feature2', 'feature3', 'feature4'
]
feature_choice = list(itertools.combinations(feature_avaliable, 2))
for i in range(len(feature_choice)):
tpr_list_temp = []
fpr_list_temp = []
BER_list_temp = []
f1_score_list_temp = []
for train_index, test_index in skf.split(training_feature,
training_label['label']):
X_train, X_val = training_feature.loc[
train_index], training_feature.loc[test_index]
y_train, y_val = training_label.loc[
train_index], training_label.loc[test_index]
X_train = X_train.loc[:,
[feature_choice[i][0], feature_choice[i][1]]]
X_val = X_val.loc[:, [feature_choice[i][0], feature_choice[i][1]]]
dis_1, dis_2 = nn_distance_calculate(X_val, X_train, y_train)
y_pred_temp = nn_predict(dis_1, dis_2, alpha, X_val)
Precall, f1_score, BER, FPR = cal_score(y_pred_temp,
y_val['label'])
tpr_list_temp.append(Precall)
fpr_list_temp.append(FPR)
BER_list_temp.append(BER)
f1_score_list_temp.append(f1_score)
alpha_list.append(alpha)
feature_0_list.append(feature_choice[i][0])
feature_1_list.append(feature_choice[i][1])
tpr_list.append(sum(tpr_list_temp) / len(tpr_list_temp))
fpr_list.append(sum(fpr_list_temp) / len(fpr_list_temp))
BER_list.append(sum(BER_list_temp) / len(BER_list_temp))
f1_score_list.append(sum(f1_score_list_temp) / len(f1_score_list_temp))
alpha = alpha + 0.001
result = {
"alpha": alpha_list,
"feature_0": feature_0_list,
"feature_1": feature_1_list,
"TPR": tpr_list,
"FPR": fpr_list,
"f1_score": f1_score_list,
"BER": BER_list
}
columns = [
"alpha", "feature_0", "feature_1", "f1_score", "TPR", "FPR", "BER"
]
result = pd.DataFrame(data=result, columns=columns)
return result
def kNN_data_ratio_adjust (X_train,y_train,X_val,y_val,k_value):
start=time.time()
label_1_amount = 40000
label_1_amount_list = []
label_2_amount_list = []
ratio_list = []
tpr_list = []
fpr_list = []
BER_list = []
f1_score_list = []
time_list=[]
while label_1_amount > 9000:
count=0
tpr_list_temp = []
fpr_list_temp = []
BER_list_temp = []
f1_score_list_temp = []
time_list_temp=[]
while (count<3):
train_data = {'feature0':X_train['feature0'],'feature1':X_train['feature1'],'feature2':X_train['feature2'],'feature3':X_train['feature3'],'feature4':X_train['feature4'],'label':y_train['label']}
train_data = pd.DataFrame(data=train_data)
Class1_sample =pd.DataFrame.sample(train_data[train_data['label']==1],label_1_amount)
Class2_sample =pd.DataFrame.sample(train_data[train_data['label']==2],20000)
res = [Class1_sample, Class2_sample]
train_com = pd.concat(res)
#print(train_com)
sample_label = pd.DataFrame(train_com['label'])
sample_feature=train_com.drop(["label"],axis=1)
start1=time.time()
neigh = KNeighborsClassifier(n_neighbors=k_value,algorithm='auto',weights ='distance')
neigh.fit(sample_feature, sample_label['label'])
y_pred = neigh.predict(X_val)
y_pred = {"label_pred":y_pred}
y_pred = pd.DataFrame(data=y_pred)
Precall,f1_score,BER,FPR = cal_score (y_pred,y_val['label'])
tpr_list_temp.append(Precall)
fpr_list_temp.append(FPR)
BER_list_temp.append(BER)
f1_score_list_temp.append(f1_score)
temp=(time.time()-start1)/60
time_list_temp.append(temp)
count=count+1
label_1_amount_list.append(label_1_amount)
label_2_amount_list.append(20000)
ratio_list.append(label_1_amount/20000)
tpr_list.append(sum(tpr_list_temp)/len(tpr_list_temp))
fpr_list.append(sum(fpr_list_temp)/len(fpr_list_temp))
BER_list.append(sum(BER_list_temp)/len(BER_list_temp))
f1_score_list.append(sum(f1_score_list_temp)/len(f1_score_list_temp))
time_list.append(sum(time_list_temp)/len(time_list_temp))
#print("current data labe 1 size:%d ,fit time:%5.1fminute"%(t,(time.time()-start1)/60))
label_1_amount = label_1_amount-2000
print("the total executing time:%5.1fminute"%((time.time()-start)/60))
result = {"label_1_amount":label_1_amount_list,"label_2_amount":label_2_amount_list,"label 1: label 2 ratio":ratio_list,"TPR":tpr_list,"FPR":fpr_list,"f1_score":f1_score_list,"BER":BER_list,"time":time_list}
columns = ["label_1_amount","label_2_amount","label 1: label 2 ratio","f1_score","TPR","FPR","BER","time"]
result = pd.DataFrame (data=result,columns=columns)
return result