Upload New File

1 files changed, 155 insertions, 0 deletions

diff --git a/attack/attackML.py b/attack/attackML.py
new file mode 100644
index 0000000..7f82372
--- /dev/null
+++ b/attack/attackML.py
@@ -0,0 +1,155 @@
+from attack.adversarialDataset import AdversarialC2Data
+from torch.utils.data import DataLoader
+from TargetModel.TargetLR import TargetLR
+from TargetModel.TargetMLP import TargetMLP
+from TargetModel.TargetSVM import TargetSVM
+from TargetModel.TargetDT import TargetDT
+from TargetModel.TargetIF import TargetIF
+from sklearn.metrics import precision_score, recall_score, f1_score
+from sklearn.metrics import confusion_matrix
+
+def attack(arch, name, theta, gamma, modify_times, target_class, model, filter_name):
+    advmodel_filename = "/home/sunhanwu/work2021/TrafficAdversarial/experiment/src/modelFile/target_cicids_{}_{}.pkt".format(arch, name)
+    advdata_filename = "/home/sunhanwu/work2021/TrafficAdversarial/experiment/src/adversarialData/advdata_cicids_{}_{}_{}_{}_{}_{}.npy".format("dnn",
+                        name, theta, gamma, modify_times, filter_name)
+    adv_data = AdversarialC2Data(advdata_filename, target_class=target_class, keep_target=True, norm=False)
+    model.load(advmodel_filename)
+    dataloader = DataLoader(adv_data, batch_size=128, shuffle=True)
+    y_pred, y_true = model.eval(dataloader)
+    precision = precision_score(y_true=y_true, y_pred=y_pred)
+    recall = recall_score(y_true=y_true, y_pred=y_pred)
+    f1 = f1_score(y_true=y_true, y_pred=y_pred)
+    con_matrix = confusion_matrix(y_true, y_pred)
+    print("{} pre: {:.2%}, recall: {:.2%}, f1: {:.2%}".format(name, precision, recall, f1))
+    print(con_matrix)
+
+def attackMTA(arch, name, theta, gamma, modify_times, target_class, model, feature_type):
+    advmodel_filename = "/home/sunhanwu/work2021/TrafficAdversarial/experiment/src/modelFile/target_mta_{}_{}_{}.pkt".format(feature_type, arch, name)
+    print("target model: {}".format(advmodel_filename))
+    advdata_filename = "/home/sunhanwu/work2021/TrafficAdversarial/experiment/src/adversarialData/advdata_mta_{}_{}_{}_{}_{}_{}.npy".format(
+        feature_type, "dnn", name, theta, gamma, modify_times)
+    adv_data = AdversarialC2Data(advdata_filename, target_class=target_class, keep_target=True, norm=False)
+    model.load(advmodel_filename)
+    dataloader = DataLoader(adv_data, batch_size=128, shuffle=True)
+    y_pred, y_true = model.eval(dataloader)
+    precision = precision_score(y_true=y_true, y_pred=y_pred)
+    recall = recall_score(y_true=y_true, y_pred=y_pred)
+    f1 = f1_score(y_true=y_true, y_pred=y_pred)
+    con_matrix = confusion_matrix(y_true, y_pred)
+    print("{} pre: {:.2%}, recall: {:.2%}, f1: {:.2%}".format(name, precision, recall, f1))
+    print(con_matrix)
+
+def attackMTACICFLowMeter(arch, name, theta, gamma, modify_times, target_class, model, proxy_arch):
+    advmodel_filename = "/home/sunhanwu/work2021/TrafficAdversarial/experiment/src/modelFile/target_mta_cicflowmeter_{}_{}.pkt".format(arch, name)
+    print("target model: {}".format(advmodel_filename))
+    advdata_filename = "/home/sunhanwu/work2021/TrafficAdversarial/experiment/src/adversarialData/advdata_mta_cicflowmeter_{}_{}_{}_{}_{}.npy".format(
+        proxy_arch, name, theta, gamma, modify_times)
+    adv_data = AdversarialC2Data(advdata_filename, target_class=target_class, keep_target=True, norm=False)
+    model.load(advmodel_filename)
+    dataloader = DataLoader(adv_data, batch_size=128, shuffle=True)
+    y_pred, y_true = model.eval(dataloader)
+    precision = precision_score(y_true=y_true, y_pred=y_pred)
+    recall = recall_score(y_true=y_true, y_pred=y_pred)
+    f1 = f1_score(y_true=y_true, y_pred=y_pred)
+    con_matrix = confusion_matrix(y_true, y_pred)
+    print("{} pre: {:.2%}, recall: {:.2%}, f1: {:.2%}".format(name, precision, recall, f1))
+    print(con_matrix)
+
+def attackCICIDSModels():
+    theta = 1
+    gamma = 0.1
+    modify_times = 20
+    target_class = 0
+    param_lr = {
+        'C': 0.3
+    }
+    param_svm = {
+        'kernel': 'rbf',
+
+    }
+    param_if= {
+        'outliers_fraction1': 0.2,
+        "n_estimators": 200
+    }
+    param_mlp = {
+        'activate': 'relu',
+        'hidden_size': (50, 25, 13),
+        'learning_rate_init': 0.001,
+        'max_iter': 200,
+        'momentum': 0.9,
+        'solver': 'adam',
+        'alpha': 0.01,
+        'batch_size': 128
+    }
+    malwares = [
+        "Botnet",
+        "Fuzzing",
+        "PortScan",
+        "BruteForce",
+        "DDoS"
+    ]
+    name = "DDoS"
+    arch = "if"
+    filer_name = "both"
+    lr = TargetLR(param_lr)
+    dt = TargetDT()
+    svm = TargetSVM(param_svm)
+    IF = TargetIF(param_if)
+    mlp = TargetMLP(param_mlp)
+    attack(arch,name,theta,gamma,modify_times,target_class,IF, filter_name=filer_name)
+
+def attackMTAModels():
+    theta = 10.0
+    gamma = 0.9
+    modify_times = 20
+    target_class = 0
+    param_lr = {
+        'c': 0.3
+    }
+    param_svm = {
+        'kernel': 'rbf',
+
+    }
+    param_if= {
+        'outliers_fraction1': 0.2,
+        "n_estimators": 200
+    }
+    param_mlp = {
+        'activate': 'relu',
+        'hidden_size': (50, 25, 13),
+        'learning_rate_init': 0.001,
+        'max_iter': 200,
+        'momentum': 0.9,
+        'solver': 'adam',
+        'alpha': 0.01,
+        'batch_size': 128
+    }
+    malwares = [
+        "Botnet",
+        "Fuzzing",
+        "PortScan",
+        "BruteForce",
+        "DDoS"
+    ]
+    Botnets = [
+        "Tofsee",
+        "Dridex",
+        "Quakbot",
+        "TrickBot",
+        "Gozi"
+    ]
+    lr = TargetLR(param_lr)
+    dt = TargetDT()
+    svm = TargetSVM(param_svm)
+    IF = TargetIF(param_if)
+    mlp = TargetMLP(param_mlp)
+    model = [lr, dt, svm, IF, mlp]
+    model_s = ['lr', 'dt', 'svm', 'if', 'mlp']
+    proxy_arch = "dnn"
+    for i in range(5):
+        for botname in Botnets:
+            print("{}:{}".format(model_s[i], botname))
+            attackMTACICFLowMeter(model_s[i],botname,theta,gamma,modify_times,target_class,model[i], proxy_arch)
+
+if __name__ == '__main__':
+    attackMTAModels()
\ No newline at end of file