z[ii] = initial_sample[ii] data[k, :] = z file = 'High_IR_Data/shuttle-2_vs_5.dat' name = file.split('.')[0] print(name) RD.Initialize_Data(file) print('Number of Positive: ', RD.Num_positive) print('Number of Negative: ', RD.Num_negative) nominal_feature = [] data = RD.get_feature() num_samples = data.shape[0] num_features = data.shape[1] num_bins = 100 bounds = np.zeros((num_bins + 1, num_features)) for i in range(num_features): if i not in nominal_feature: bounds[:, i] = np.histogram(data[:, i], bins=num_bins)[1] nf = RD.get_negative_feature() transfer(nf, bounds, num_features, nominal_feature) pf = RD.get_positive_feature() transfer(pf, bounds, num_features, nominal_feature) bayes = BayesianNetwork.from_samples(nf, algorithm='chow-liu') pt = bayes.log_probability(nf).sum()
Num_Cross_Folders = 5 G_Mean = np.linspace(0, 0, Num_Cross_Folders) Sensitivity = np.linspace(0, 0, Num_Cross_Folders) Specificity = np.linspace(0, 0, Num_Cross_Folders) G_Mean_GAN = np.linspace(0, 0, Num_Cross_Folders) Sensitivity_GAN = np.linspace(0, 0, Num_Cross_Folders) Specificity_GAN = np.linspace(0, 0, Num_Cross_Folders) for j in range(Num_Cross_Folders): # dir_train = "glass1-5-fold/glass1-5-" + str(j+1) + "tra.dat" # dir_test = "glass1-5-fold/glass1-5-" + str(j+1) + "tst.dat" dir_train = "page-blocks0-5-fold/page-blocks0-5-" + str(j + 1) + "tra.dat" dir_test = "page-blocks0-5-fold/page-blocks0-5-" + str(j + 1) + "tst.dat" RD.Initialize_Data(dir_train) Train_Feature = RD.get_feature() Train_Label = RD.get_label() Train_Label = Train_Label.ravel() print(Train_Feature.shape) print(Train_Label.size) # clf = svm.SVC(C=1, kernel='rbf', gamma= 0.2) # clf.fit(Train_Feature, Train_Label) Feature_samples = RD.get_positive_feature() G = GAN_Build(Feature_samples) Sudo_Samples = Over_Sampling(G, RD.Num_negative - RD.Num_positive, 6) print(Sudo_Samples[0]) print(Sudo_Samples[-1]) Train_Feature = np.concatenate((Train_Feature, Sudo_Samples)) print(Train_Feature[0])