import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import random
from sklearn.neighbors import NearestNeighbors
from scipy import ndarray
filename='preprocessed_majority.csv'
datapd_0=pd.read_csv(filename, index_col=0)
filename='preprocessed_minority.csv'
datapd_1=pd.read_csv(filename, index_col=0 )
print('Majority class dataframe shape:', datapd_0.shape)
print('Minority class dataframe shape:', datapd_1.shape)
n_feat=datapd_0.shape[1]
print('Imbalance Ratio:', datapd_0.shape[0]/datapd_1.shape[0])
features_0=np.asarray(datapd_0)
features_1=np.asarray(datapd_1)
s=42
features_1=np.take(features_1,np.random.RandomState(seed=s).permutation(features_1.shape[0]),axis=0,out=features_1)
features_0=np.take(features_0,np.random.RandomState(seed=s).permutation(features_0.shape[0]),axis=0,out=features_0)
a=len(features_1)//3
b=len(features_0)//3
fold_1_min=features_1[0:a]
fold_1_maj=features_0[0:b]
fold_1_tst=np.concatenate((fold_1_min,fold_1_maj))
lab_1_tst=np.concatenate((np.zeros(len(fold_1_min))+1, np.zeros(len(fold_1_maj))))
fold_2_min=features_1[a:2*a]
fold_2_maj=features_0[b:2*b]
fold_2_tst=np.concatenate((fold_2_min,fold_2_maj))
lab_2_tst=np.concatenate((np.zeros(len(fold_1_min))+1, np.zeros(len(fold_1_maj))))
fold_3_min=features_1[2*a:]
fold_3_maj=features_0[2*b:]
fold_3_tst=np.concatenate((fold_3_min,fold_3_maj))
lab_3_tst=np.concatenate((np.zeros(len(fold_3_min))+1, np.zeros(len(fold_3_maj))))
fold_1_trn=np.concatenate((fold_2_min,fold_3_min,fold_2_maj,fold_3_maj))
lab_1_trn=np.concatenate((np.zeros(a+len(fold_3_min))+1,np.zeros(b+len(fold_3_maj))))
fold_2_trn=np.concatenate((fold_1_min,fold_3_min,fold_1_maj,fold_3_maj))
lab_2_trn=np.concatenate((np.zeros(a+len(fold_3_min))+1,np.zeros(b+len(fold_3_maj))))
fold_3_trn=np.concatenate((fold_2_min,fold_1_min,fold_2_maj,fold_1_maj))
lab_3_trn=np.concatenate((np.zeros(2*a)+1,np.zeros(2*b)))
training_folds_feats=[fold_1_trn,fold_2_trn,fold_3_trn]
testing_folds_feats=[fold_1_tst,fold_2_tst,fold_3_tst]
training_folds_labels=[lab_1_trn,lab_2_trn,lab_3_trn]
testing_folds_labels=[lab_1_tst,lab_2_tst,lab_3_tst]
def lr(X_train,y_train,X_test,y_test):
from sklearn import metrics
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import confusion_matrix
from sklearn.metrics import average_precision_score
from sklearn.metrics import f1_score
from sklearn.metrics import balanced_accuracy_score
logreg = LogisticRegression(C=1e5, solver='lbfgs', multi_class='multinomial', class_weight={0: 1, 1: 1})
logreg.fit(X_train, y_train)
y_pred= logreg.predict(X_test)
con_mat=confusion_matrix(y_test,y_pred)
bal_acc=balanced_accuracy_score(y_test,y_pred)
tn, fp, fn, tp = con_mat.ravel()
print('tn, fp, fn, tp:', tn, fp, fn, tp)
f1 = f1_score(y_test, y_pred)
print('balanced accuracy_LR:', bal_acc)
print('f1 score_LR:', f1)
print('confusion matrix_LR',con_mat)
return(f1, bal_acc ,con_mat)
def svm(X_train,y_train,X_test,y_test):
from sklearn import preprocessing
from sklearn import metrics
#from sklearn import svm
from sklearn.svm import LinearSVC
from sklearn.metrics import confusion_matrix
from sklearn.metrics import f1_score
from sklearn.metrics import average_precision_score
from sklearn.metrics import balanced_accuracy_score
X_train = preprocessing.scale(X_train)
X_test = preprocessing.scale(X_test)
#svm= svm.SVC(kernel='linear', decision_function_shape='ovo', class_weight={0: 1., 1: 1.},probability=True)
svm= LinearSVC(random_state=0, tol=1e-5)
svm.fit(X_train, y_train)
y_pred= svm.predict(X_test)
con_mat=confusion_matrix(y_test,y_pred)
bal_acc=balanced_accuracy_score(y_test,y_pred)
tn, fp, fn, tp = con_mat.ravel()
print('tn, fp, fn, tp:', tn, fp, fn, tp)
f1 = f1_score(y_test, y_pred)
print('balanced accuracy_SVM:', bal_acc)
print('f1 score_SVM:', f1)
print('confusion matrix_SVM',con_mat)
return( f1, bal_acc ,con_mat)
def knn(X_train,y_train,X_test,y_test):
from sklearn import metrics
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import confusion_matrix
from sklearn.metrics import f1_score
from sklearn.metrics import average_precision_score
from sklearn.metrics import balanced_accuracy_score
knn = KNeighborsClassifier(n_neighbors=10)
knn.fit(X_train, y_train)
y_pred= knn.predict(X_test)
con_mat=confusion_matrix(y_test,y_pred)
bal_acc=balanced_accuracy_score(y_test,y_pred)
tn, fp, fn, tp = con_mat.ravel()
print('tn, fp, fn, tp:', tn, fp, fn, tp)
print('balanced accuracy_KNN:', bal_acc)
f1 = f1_score(y_test, y_pred)
print('f1 score_KNN:', f1)
print('confusion matrix_KNN',con_mat)
return(f1, bal_acc, con_mat)
def Neb_grps(data,near_neb):
nbrs = NearestNeighbors(n_neighbors=near_neb, algorithm='ball_tree').fit(data)
distances, indices = nbrs.kneighbors(data)
neb_class=[]
for i in (indices):
neb_class.append(i)
return(np.asarray(neb_class))
def LoRAS(data,num_samples,shadow,sigma,num_RACOS,num_afcomb):
np.random.seed(42)
data_shadow=([])
for i in range (num_samples):
c=0
while c<shadow:
data_shadow.append(data[i]+np.random.normal(0,sigma))
c=c+1
data_shadow==np.asarray(data_shadow)
data_shadow_lc=([])
for i in range(num_RACOS):
idx = np.random.randint(shadow*num_samples, size=num_afcomb)
w=np.random.randint(100, size=len(idx))
aff_w=np.asarray(w/sum(w))
data_tsl=np.array(data_shadow)[idx,:]
data_tsl_=np.dot(aff_w, data_tsl)
data_shadow_lc.append(data_tsl_)
return(np.asarray(data_shadow_lc))
def LoRAS_gen(num_samples,shadow,sigma,num_RACOS,num_afcomb):
RACOS_set=[]
for i in range (len(nb_list)):
RACOS_i= LoRAS(features_1_trn[nb_list[i]],num_samples,shadow,sigma,num_RACOS,num_afcomb)
RACOS_set.append(RACOS_i)
LoRAS_set=np.asarray(RACOS_set)
LoRAS_1=np.reshape(LoRAS_set,(len(features_1_trn)*num_RACOS,n_feat))
return(np.concatenate((LoRAS_1,features_1_trn)))
def OVS(training_data,training_labels,neb):
from imblearn.over_sampling import SMOTE
sm = SMOTE(random_state=62, k_neighbors=neb, kind='regular',ratio=1)
SMOTE_feat, SMOTE_labels = sm.fit_resample(training_data,training_labels)
smbl1 = SMOTE(random_state=62, k_neighbors=neb, kind='borderline1',ratio=1)
SMOTE_feat_bl1, SMOTE_labels_bl1 = smbl1.fit_resample(training_data,training_labels)
smbl2 = SMOTE(random_state=62, k_neighbors=neb, kind='borderline2',ratio=1)
SMOTE_feat_bl2, SMOTE_labels_bl2 = smbl2.fit_resample(training_data,training_labels)
smsvm = SMOTE(random_state=62, k_neighbors=neb, kind='svm',ratio=1)
SMOTE_feat_svm, SMOTE_labels_svm = smsvm.fit_resample(training_data,training_labels)
from imblearn.over_sampling import ADASYN
ad = ADASYN(random_state=62,n_neighbors=neb, ratio=1)
ADASYN_feat, ADASYN_labels = ad.fit_resample(training_data,training_labels)
return(SMOTE_feat, SMOTE_labels,SMOTE_feat_bl1, SMOTE_labels_bl1, SMOTE_feat_bl2, SMOTE_labels_bl2,SMOTE_feat_svm, SMOTE_labels_svm,ADASYN_feat, ADASYN_labels)
LR=[]
SVM=[]
KNN=[]
LR_SM=[]
SVM_SM=[]
KNN_SM=[]
LR_SMBL1=[]
SVM_SMBL1=[]
KNN_SMBL1=[]
LR_SMBL2=[]
SVM_SMBL2=[]
KNN_SMBL2=[]
LR_SMSVM=[]
SVM_SMSVM=[]
KNN_SMSVM=[]
LR_ADA=[]
SVM_ADA=[]
KNN_ADA=[]
i=0
while i<3:
SMOTE_feat, SMOTE_labels,SMOTE_feat_bl1, SMOTE_labels_bl1, SMOTE_feat_bl2, SMOTE_labels_bl2,SMOTE_feat_svm, SMOTE_labels_svm,ADASYN_feat, ADASYN_labels=OVS(training_folds_feats[i],training_folds_labels[i],3)
f1_lr, bal_acc_lr, mat_lr=lr(training_folds_feats[i],training_folds_labels[i],testing_folds_feats[i],testing_folds_labels[i])
LR.append([f1_lr, bal_acc_lr])
f1_svm,bal_acc_svm,mat_svm=svm(training_folds_feats[i],training_folds_labels[i],testing_folds_feats[i],testing_folds_labels[i])
SVM.append([f1_svm,bal_acc_svm])
f1_knn,bal_acc_knn,mat_knn=knn(training_folds_feats[i],training_folds_labels[i],testing_folds_feats[i],testing_folds_labels[i])
KNN.append([f1_knn,bal_acc_knn])
f1_lr_SMOTE,bal_acc_lr_SMOTE,mat_lr_SMOTE=lr(SMOTE_feat,SMOTE_labels,testing_folds_feats[i],testing_folds_labels[i])
LR_SM.append([f1_lr_SMOTE,bal_acc_lr_SMOTE])
f1_svm_SMOTE,bal_acc_svm_SMOTE,mat_svm_SMOTE=svm(SMOTE_feat,SMOTE_labels,testing_folds_feats[i],testing_folds_labels[i])
SVM_SM.append([f1_svm_SMOTE,bal_acc_svm_SMOTE])
f1_knn_SMOTE,bal_acc_knn_SMOTE,mat_knn_SMOTE=knn(SMOTE_feat,SMOTE_labels,testing_folds_feats[i],testing_folds_labels[i])
KNN_SM.append([f1_knn_SMOTE,bal_acc_knn_SMOTE])
f1_lr_SMOTE_bl1,bal_acc_lr_SMOTE_bl1,mat_lr_SMOTE_bl1=lr(SMOTE_feat_bl1,SMOTE_labels_bl1,testing_folds_feats[i],testing_folds_labels[i])
LR_SMBL1.append([f1_lr_SMOTE_bl1,bal_acc_lr_SMOTE_bl1])
f1_svm_SMOTE_bl1,bal_acc_svm_SMOTE_bl1,mat_svm_SMOTE_bl1=svm(SMOTE_feat_bl1,SMOTE_labels_bl1,testing_folds_feats[i],testing_folds_labels[i])
SVM_SMBL1.append([f1_svm_SMOTE_bl1,bal_acc_svm_SMOTE_bl1])
f1_knn_SMOTE_bl1,bal_acc_knn_SMOTE_bl1,mat_knn_SMOTE_bl1=knn(SMOTE_feat_bl1,SMOTE_labels_bl1,testing_folds_feats[i],testing_folds_labels[i])
KNN_SMBL1.append([f1_knn_SMOTE_bl1,bal_acc_knn_SMOTE_bl1])
f1_lr_SMOTE_bl2,bal_acc_lr_SMOTE_bl2,mat_lr_SMOTE_bl2=lr(SMOTE_feat_bl2,SMOTE_labels_bl2,testing_folds_feats[i],testing_folds_labels[i])
LR_SMBL2.append([f1_lr_SMOTE_bl2,bal_acc_lr_SMOTE_bl2])
f1_svm_SMOTE_bl2,bal_acc_svm_SMOTE_bl2,mat_svm_SMOTE_bl2=svm(SMOTE_feat_bl1,SMOTE_labels_bl1,testing_folds_feats[i],testing_folds_labels[i])
SVM_SMBL2.append([f1_svm_SMOTE_bl2,bal_acc_svm_SMOTE_bl2])
f1_knn_SMOTE_bl2,bal_acc_knn_SMOTE_bl2,mat_knn_SMOTE_bl2=knn(SMOTE_feat_bl2,SMOTE_labels_bl2,testing_folds_feats[i],testing_folds_labels[i])
KNN_SMBL2.append([f1_knn_SMOTE_bl2,bal_acc_knn_SMOTE_bl2])
f1_lr_SMOTE_svm,bal_acc_lr_SMOTE_svm,mat_lr_SMOTE_svm=lr(SMOTE_feat_svm,SMOTE_labels_svm,testing_folds_feats[i],testing_folds_labels[i])
LR_SMSVM.append([f1_lr_SMOTE_svm,bal_acc_lr_SMOTE_svm])
f1_svm_SMOTE_svm,bal_acc_svm_SMOTE_svm,mat_svm_SMOTE_svm=svm(SMOTE_feat_svm,SMOTE_labels_svm,testing_folds_feats[i],testing_folds_labels[i])
SVM_SMSVM.append([f1_svm_SMOTE_svm,bal_acc_svm_SMOTE_svm])
f1_knn_SMOTE_svm,bal_acc_knn_SMOTE_svm,mat_knn_SMOTE_svm=knn(SMOTE_feat_svm,SMOTE_labels_svm,testing_folds_feats[i],testing_folds_labels[i])
KNN_SMSVM.append([f1_knn_SMOTE_svm,bal_acc_knn_SMOTE_svm])
f1_lr_ADASYN,bal_acc_lr_ADASYN,mat_lr_ADASYN=lr(ADASYN_feat,ADASYN_labels,testing_folds_feats[i],testing_folds_labels[i])
LR_ADA.append([f1_lr_ADASYN,bal_acc_lr_ADASYN])
f1_svm_ADASYN,bal_acc_svm_ADASYN,mat_svm_ADASYN=svm(ADASYN_feat,ADASYN_labels,testing_folds_feats[i],testing_folds_labels[i])
SVM_ADA.append([f1_svm_ADASYN,bal_acc_svm_ADASYN])
f1_knn_ADASYN,bal_acc_knn_ADASYN,mat_knn_ADASYN=knn(ADASYN_feat,ADASYN_labels,testing_folds_feats[i],testing_folds_labels[i])
KNN_ADA.append([f1_knn_ADASYN,bal_acc_knn_ADASYN])
i=i+1
LR_LoRAS=[]
SVM_LoRAS=[]
KNN_LoRAS=[]
for i in range(3):
features = training_folds_feats[i]
labels= training_folds_labels[i]
label_1=np.where(labels == 1)[0]
label_1=list(label_1)
features_1_trn=features[label_1]
label_0=np.where(labels == 0)[0]
label_0=list(label_0)
features_0_trn=features[label_0]
num_samples=3
shadow=100
sigma=.005
num_RACOS=(len(features_0_trn)-len(features_1_trn))//len(features_1_trn)
num_afcomb=50
nb_list=Neb_grps(features_1_trn, num_samples)
LoRAS_1=LoRAS_gen(num_samples,shadow,sigma,num_RACOS,num_afcomb)
LoRAS_train=np.concatenate((LoRAS_1,features_0_trn))
LoRAS_labels=np.concatenate((np.zeros(len(LoRAS_1))+1, np.zeros(len(features_0_trn))))
f1_lr_LoRAS,bal_acc_lr_LoRAS,mat_lr_LoRAS=lr(LoRAS_train,LoRAS_labels,testing_folds_feats[i],testing_folds_labels[i])
LR_LoRAS.append([f1_lr_LoRAS,bal_acc_lr_LoRAS])
f1_svm_LoRAS,bal_acc_svm_LoRAS,mat_svm_LoRAS=svm(LoRAS_train,LoRAS_labels,testing_folds_feats[i],testing_folds_labels[i])
SVM_LoRAS.append([f1_svm_LoRAS,bal_acc_svm_LoRAS])
f1_knn_LoRAS,bal_acc_knn_LoRAS,mat_knn_LoRAS=knn(LoRAS_train,LoRAS_labels,testing_folds_feats[i],testing_folds_labels[i])
KNN_LoRAS.append([f1_knn_LoRAS,bal_acc_knn_LoRAS])
LR_tLoRAS=[]
SVM_tLoRAS=[]
KNN_tLoRAS=[]
from sklearn.manifold import TSNE
for i in range(3):
features = training_folds_feats[i]
labels= training_folds_labels[i]
label_1=np.where(labels == 1)[0]
label_1=list(label_1)
features_1_trn=features[label_1]
label_0=np.where(labels == 0)[0]
label_0=list(label_0)
features_0_trn=features[label_0]
data_embedded_min = TSNE().fit_transform(features_1_trn)
result_min= pd.DataFrame(data = data_embedded_min, columns = ['t-SNE0', 't-SNE1'])
min_t=np.asmatrix(result_min)
min_t=min_t[0:len(features_1_trn)]
min_t=min_t[:, [0,1]]
num_samples=3
shadow=100
sigma=.005
num_RACOS=(len(features_0_trn)-len(features_1_trn))//len(features_1_trn)
num_afcomb=50
nb_list=Neb_grps(min_t, num_samples)
LoRAS_1=LoRAS_gen(num_samples,shadow,sigma,num_RACOS,num_afcomb)
LoRAS_train=np.concatenate((LoRAS_1,features_0_trn))
LoRAS_labels=np.concatenate((np.zeros(len(LoRAS_1))+1, np.zeros(len(features_0_trn))))
f1_lr_LoRAS,bal_acc_lr_LoRAS,mat_lr_LoRAS=lr(LoRAS_train,LoRAS_labels,testing_folds_feats[i],testing_folds_labels[i])
LR_tLoRAS.append([f1_lr_LoRAS,bal_acc_lr_LoRAS])
f1_svm_LoRAS,bal_acc_svm_LoRAS,mat_svm_LoRAS=svm(LoRAS_train,LoRAS_labels,testing_folds_feats[i],testing_folds_labels[i])
SVM_tLoRAS.append([f1_svm_LoRAS,bal_acc_svm_LoRAS])
f1_knn_LoRAS,bal_acc_knn_LoRAS,mat_knn_LoRAS=knn(LoRAS_train,LoRAS_labels,testing_folds_feats[i],testing_folds_labels[i])
KNN_tLoRAS.append([f1_knn_LoRAS,bal_acc_knn_LoRAS])
def stats(arr):
x=np.mean(np.asarray(arr), axis = 0)
y=np.std(np.asarray(arr), axis = 0)
return(x,y)
print('F1|Balanced Accuracy :: mean|sd')
print('Without Oversampling')
LR_m, LR_sd=stats(LR)
print('lr:',LR_m, LR_sd)
SVM_m, SVM_sd=stats(SVM)
print('svm:',SVM_m, SVM_sd)
KNN_m, KNN_sd= stats(KNN)
print('knn:',KNN_m, KNN_sd)
print('SMOTE Oversampling')
LR_SM_m, LR_SM_sd=stats(LR_SM)
print('lr:',LR_SM_m, LR_SM_sd)
SVM_SM_m, SVM_SM_sd=stats(SVM_SM)
print('svm:',SVM_SM_m, SVM_SM_sd)
KNN_SM_m, KNN_SM_sd=stats(KNN_SM)
print('knn:',KNN_SM_m, KNN_SM_sd)
print('SMOTE-Bl1 Oversampling')
LR_SMBL1_m, LR_SMBL1_sd=stats(LR_SMBL1)
print('lr:',LR_SMBL1_m, LR_SMBL1_sd)
SVM_SMBL1_m,SVM_SMBL1_sd=stats(SVM_SMBL1)
print('svm:',SVM_SMBL1_m,SVM_SMBL1_sd)
KNN_SMBL1_m, KNN_SMBL1_sd= stats(KNN_SMBL1)
print('knn:',KNN_SMBL1_m, KNN_SMBL1_sd)
print('SMOTE-Bl2 Oversampling')
LR_SMBL2_m, LR_SMBL2_sd=stats(LR_SMBL2)
print('lr:',LR_SMBL2_m, LR_SMBL2_sd)
SVM_SMBL2_m, SVM_SMBL2_sd=stats(SVM_SMBL2)
print('svm:',SVM_SMBL2_m, SVM_SMBL2_sd)
KNN_SMBL2_m, KNN_SMBL2_sd= stats(KNN_SMBL2)
print('knn:',KNN_SMBL2_m, KNN_SMBL2_sd)
print('SMOTE-SVM Oversampling')
LR_SMSVM_m, LR_SMSVM_sd=stats(LR_SMSVM)
print('lr:',LR_SMSVM_m, LR_SMSVM_sd)
SVM_SMSVM_m, SVM_SMSVM_sd=stats(SVM_SMSVM)
print('svm:',SVM_SMSVM_m, SVM_SMSVM_sd)
KNN_SMSVM_m, KNN_SMSVM_sd= stats(KNN_SMSVM)
print('knn:',KNN_SMSVM_m, KNN_SMSVM_sd)
print('ADASYN Oversampling')
LR_ADA_m, LR_ADA_sd=stats(LR_ADA)
print('lr:',LR_ADA_m, LR_ADA_sd)
SVM_ADA_m, SVM_ADA_sd=stats(SVM_ADA)
print('svm:',SVM_ADA_m, SVM_ADA_sd)
KNN_ADA_m, KNN_ADA_sd=stats(KNN_ADA)
print('knn:',KNN_ADA_m, KNN_ADA_sd)
print('LoRAS Oversampling')
LR_LoRAS_m, LR_LoRAS_sd=stats(LR_LoRAS)
print('lr:',LR_LoRAS_m, LR_LoRAS_sd)
SVM_LoRAS_m, SVM_LoRAS_sd=stats(SVM_LoRAS)
print('svm:',SVM_LoRAS_m, SVM_LoRAS_sd)
KNN_LoRAS_m, KNN_LoRAS_sd=stats(KNN_LoRAS)
print('knn:',KNN_LoRAS_m, KNN_LoRAS_sd)
print('tLoRAS Oversampling')
LR_tLoRAS_m, LR_tLoRAS_sd=stats(LR_tLoRAS)
print('lr:',LR_tLoRAS_m, LR_tLoRAS_sd)
SVM_tLoRAS_m, SVM_tLoRAS_sd=stats(SVM_tLoRAS)
print('svm:',SVM_tLoRAS_m, SVM_tLoRAS_sd)
KNN_tLoRAS_m, KNN_tLoRAS_sd=stats(KNN_tLoRAS)
print('knn:',KNN_tLoRAS_m, KNN_tLoRAS_sd)
!jupyter nbconvert --to html CV.ipynb