In [1]:
import pandas as pd
import numpy as np
import os
import time
from collections import namedtuple
pd.set_option("display.max_rows",35)
%matplotlib inline
In [2]:
%%bash
rm dataset/scores/tf_dense_only_nsl_kdd_scores_all.pkl
In [3]:
class dataset:
kdd_train_2labels = pd.read_pickle("dataset/kdd_train_2labels.pkl")
kdd_test_2labels = pd.read_pickle("dataset/kdd_test_2labels.pkl")
kdd_test__2labels = pd.read_pickle("dataset/kdd_test__2labels.pkl")
kdd_train_5labels = pd.read_pickle("dataset/kdd_train_5labels.pkl")
kdd_test_5labels = pd.read_pickle("dataset/kdd_test_5labels.pkl")
In [4]:
dataset.kdd_train_2labels.shape
Out[4]:
In [5]:
dataset.kdd_test_2labels.shape
Out[5]:
In [6]:
from sklearn import model_selection as ms
from sklearn import preprocessing as pp
class preprocess:
output_columns_2labels = ['is_Normal','is_Attack']
x_input = dataset.kdd_train_2labels.drop(output_columns_2labels, axis = 1)
y_output = dataset.kdd_train_2labels.loc[:,output_columns_2labels]
x_test_input = dataset.kdd_test_2labels.drop(output_columns_2labels, axis = 1)
y_test = dataset.kdd_test_2labels.loc[:,output_columns_2labels]
x_test__input = dataset.kdd_test__2labels.drop(output_columns_2labels, axis = 1)
y_test_ = dataset.kdd_test__2labels.loc[:,output_columns_2labels]
ss = pp.StandardScaler()
x_train = ss.fit_transform(x_input)
x_test = ss.transform(x_test_input)
x_test_ = ss.transform(x_test__input)
y_train = y_output.values
y_test = y_test.values
y_test_ = y_test_.values
preprocess.x_train.std()
Out[6]:
In [7]:
import tensorflow as tf
from tensorflow.contrib.legacy_seq2seq.python.ops.seq2seq import basic_rnn_seq2seq
from tensorflow.contrib.rnn import RNNCell, LSTMCell, MultiRNNCell
In [8]:
class network(object):
input_dim = 122
classes = 2
hidden_encoder_dim = 122
hidden_layers = 1
latent_dim = 10
hidden_decoder_dim = 122
lam = 0.01
def __init__(self, classes, hidden_layers, num_of_features):
self.classes = classes
self.hidden_layers = hidden_layers
self.latent_dim = num_of_features
def build_layers(self):
tf.reset_default_graph()
#learning_rate = tf.Variable(initial_value=0.001)
input_dim = self.input_dim
classes = self.classes
hidden_encoder_dim = self.hidden_encoder_dim
hidden_layers = self.hidden_layers
latent_dim = self.latent_dim
hidden_decoder_dim = self.hidden_decoder_dim
lam = self.lam
with tf.variable_scope("Input"):
self.x_input = tf.placeholder("float", shape=[None, 1, input_dim])
self.y_input_ = tf.placeholder("float", shape=[None, 1, classes])
self.keep_prob = tf.placeholder("float")
self.lr = tf.placeholder("float")
self.x_list = tf.unstack(self.x_input, axis= 1)
self.y_list_ = tf.unstack(self.y_input_, axis = 1)
self.y_ = self.y_list_[0]
#GO = tf.fill((tf.shape(self.x)[0], 1), 0.5)
#y_with_GO = tf.stack([self.y_, GO])
with tf.variable_scope("lstm"):
multi_cell = MultiRNNCell([LSTMCell(input_dim) for i in range(hidden_layers)] )
self.y, states = basic_rnn_seq2seq(self.x_list, self.y_list_, multi_cell)
#self.y = tf.slice(self.y, [0, 0], [-1,2])
#self.out = tf.squeeze(self.y)
#self.y = tf.layers.dense(self.y[0], classes, activation = None)
self.y = tf.slice(self.y[0], [0, 0], [-1,2])
with tf.variable_scope("Loss"):
self.regularized_loss = tf.losses.mean_squared_error(self.y_, self.y)
correct_prediction = tf.equal(tf.argmax(self.y_, 1), tf.argmax(self.y, 1))
self.tf_accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32), name = "Accuracy")
with tf.variable_scope("Optimizer"):
learning_rate=self.lr
optimizer = tf.train.AdamOptimizer(learning_rate)
gradients, variables = zip(*optimizer.compute_gradients(self.regularized_loss))
gradients = [
None if gradient is None else tf.clip_by_value(gradient, -1, 1)
for gradient in gradients]
self.train_op = optimizer.apply_gradients(zip(gradients, variables))
#self.train_op = optimizer.minimize(self.regularized_loss)
# add op for merging summary
#self.summary_op = tf.summary.merge_all()
self.pred = tf.argmax(self.y, axis = 1)
self.actual = tf.argmax(self.y_, axis = 1)
# add Saver ops
self.saver = tf.train.Saver()
batch_iterations = 200
x_train, x_valid, y_train, y_valid, = ms.train_test_split(preprocess.x_train, preprocess.y_train, test_size=0.1) batch_indices = np.array_split(np.arange(x_train.shape[0]), batch_iterations)
for i in batch_indices: print(x_train[i,np.newaxis,:]) print(y_train[i,np.newaxis,:])
In [9]:
import collections
import sklearn.metrics as me
class Train:
result = namedtuple("score", ['epoch', 'no_of_features','hidden_layers','train_score', 'test_score', 'f1_score', 'test_score_20', 'f1_score_20', 'time_taken'])
predictions = {}
predictions_ = {}
results = []
best_acc = 0
best_acc_global = 0
def train(epochs, net, h,f, lrs):
batch_iterations = 200
train_loss = None
Train.best_acc = 0
os.makedirs("dataset/tf_lstm_nsl_kdd-orig/hidden layers_{}_features count_{}".format(h,f),
exist_ok = True)
with tf.Session() as sess:
#summary_writer_train = tf.summary.FileWriter('./logs/kdd/VAE/training', graph=sess.graph)
#summary_writer_valid = tf.summary.FileWriter('./logs/kdd/VAE/validation')
sess.run(tf.global_variables_initializer())
start_time = time.perf_counter()
accuracy, pred_value, actual_value, y_pred = sess.run([net.tf_accuracy,
net.pred,
net.actual, net.y],
feed_dict={net.x_input: preprocess.x_test[:,np.newaxis,:],
net.y_input_: preprocess.y_test[:,np.newaxis,:],
net.keep_prob:1})
print("Initial Accuracy, before training: {}".format(accuracy))
for c, lr in enumerate(lrs):
for epoch in range(1, (epochs+1)):
x_train, x_valid, y_train, y_valid, = ms.train_test_split(preprocess.x_train,
preprocess.y_train,
test_size=0.1)
batch_indices = np.array_split(np.arange(x_train.shape[0]),
batch_iterations)
for i in batch_indices:
_, train_loss = sess.run([net.train_op, net.regularized_loss], #net.summary_op
feed_dict={net.x_input: x_train[i,np.newaxis,:],
net.y_input_: y_train[i,np.newaxis,:],
net.keep_prob:1, net.lr:lr})
#summary_writer_train.add_summary(summary_str, epoch)
if(train_loss > 1e9):
print("Step {} | Training Loss: {:.6f}".format(epoch, train_loss))
valid_accuracy,valid_loss = sess.run([net.tf_accuracy, net.regularized_loss], #net.summary_op
feed_dict={net.x_input: x_valid[:,np.newaxis,:],
net.y_input_: y_valid[:,np.newaxis,:],
net.keep_prob:1, net.lr:lr})
#summary_writer_valid.add_summary(summary_str, epoch)
accuracy, pred_value, actual_value, y_pred = sess.run([net.tf_accuracy,
net.pred,
net.actual, net.y],
feed_dict={net.x_input: preprocess.x_test[:,np.newaxis,:],
net.y_input_: preprocess.y_test[:,np.newaxis,:],
net.keep_prob:1, net.lr:lr})
f1_score = me.f1_score(actual_value, pred_value)
accuracy_, pred_value_, actual_value_, y_pred_ = sess.run([net.tf_accuracy,
net.pred,
net.actual, net.y],
feed_dict={net.x_input: preprocess.x_test_[:,np.newaxis,:],
net.y_input_: preprocess.y_test_[:,np.newaxis,:],
net.keep_prob:1, net.lr:lr})
f1_score_ = me.f1_score(actual_value_, pred_value_)
print("Step {} | Training Loss: {:.6f} | Train Accuracy: {:.6f} | Test Accuracy: {:.6f}, {:.6f}".format(epoch, train_loss, valid_accuracy, accuracy, accuracy_))
if accuracy > Train.best_acc_global:
Train.best_acc_global = accuracy
Train.pred_value = pred_value
Train.actual_value = actual_value
Train.pred_value_ = pred_value_
Train.actual_value_ = actual_value_
Train.best_parameters = "Hidden Layers:{}, Features Count:{}".format(h, f)
if accuracy > Train.best_acc:
#net.saver.save(sess, "dataset/tf_vae_only_nsl_kdd_hidden layers_{}_features count_{}".format(epochs,h,f))
#Train.results.append(Train.result(epochs, f, h,valid_accuracy, accuracy))
#curr_pred = pd.DataFrame({"Attack_prob":y_pred[:,-2], "Normal_prob":y_pred[:, -1]})
#Train.predictions.update({"{}_{}_{}".format(epochs,f,h):curr_pred})
Train.best_acc = accuracy
if not (np.isnan(train_loss)):
net.saver.save(sess,
"dataset/tf_lstm_nsl_kdd-orig/hidden layers_{}_features count_{}/model"
.format(h,f),
global_step = epoch,
write_meta_graph=False)
curr_pred = pd.DataFrame({"Attack_prob":y_pred[:,-2], "Normal_prob":y_pred[:, -1], "Prediction":pred_value, "Actual":actual_value})
curr_pred_ = pd.DataFrame({"Attack_prob":y_pred_[:,-2], "Normal_prob":y_pred_[:, -1], "Prediction":pred_value_, "Actual": actual_value_})
Train.predictions.update({"{}_{}_{}".format((epochs+1)* (c+1),f,h):
(curr_pred,
Train.result((epochs+1)*(c+1), f, h,valid_accuracy, accuracy, f1_score, accuracy_, f1_score_, time.perf_counter() - start_time))})
Train.predictions_.update({"{}_{}_{}".format((epochs+1)* (c+1),f,h):
(curr_pred_,
Train.result((epochs+1)*(c+1), f, h,valid_accuracy, accuracy, f1_score, accuracy_, f1_score_, time.perf_counter() - start_time))})
In [10]:
import itertools
df_results = []
past_scores = []
class Hyperparameters:
# features_arr = [2, 4, 8, 16, 32, 64, 128, 256]
# hidden_layers_arr = [2, 4, 6, 10]
def start_training():
global df_results
global past_scores
Train.predictions = {}
Train.results = []
features_arr = [1] #[4, 8, 16, 32]
hidden_layers_arr = [1, 3]
epochs = [10]
lrs = [1e-2, 1e-3]
for e, h, f in itertools.product(epochs, hidden_layers_arr, features_arr):
print("Current Layer Attributes - epochs:{} hidden layers:{} features count:{}".format(e,h,f))
n = network(2,h,f)
n.build_layers()
Train.train(e, n, h,f, lrs)
dict1 = {}
dict1_ = {}
dict2 = []
for k, (v1, v2) in Train.predictions.items():
dict1.update({k: v1})
dict2.append(v2)
for k, (v1_, v2) in Train.predictions.items():
dict1_.update({k: v1_})
Train.predictions = dict1
Train.predictions_ = dict1_
Train.results = dict2
df_results = pd.DataFrame(Train.results)
temp = df_results.set_index(['no_of_features', 'hidden_layers'])
if not os.path.isfile('dataset/scores/tf_lstm_nsl_kdd-orig_all.pkl'):
past_scores = temp
else:
past_scores = pd.read_pickle("dataset/scores/tf_lstm_nsl_kdd-orig_all.pkl")
past_scores.append(temp).to_pickle("dataset/scores/tf_lstm_nsl_kdd-orig_all.pkl")
In [11]:
#%%timeit -r 10
Hyperparameters.start_training()
In [12]:
pd.Panel(Train.predictions).to_pickle("dataset/tf_lstm_nsl_kdd_predictions.pkl")
pd.Panel(Train.predictions_).to_pickle("dataset/tf_lstm_nsl_kdd_predictions__.pkl")
df_results.to_pickle("dataset/tf_lstm_nsl_kdd_scores.pkl")
In [13]:
import numpy as np
import matplotlib.pyplot as plt
import itertools
def plot_confusion_matrix(cm, classes,
normalize=False,
title='Confusion matrix',
cmap=plt.cm.Blues):
"""
This function prints and plots the confusion matrix.
Normalization can be applied by setting `normalize=True`.
"""
np.set_printoptions(precision=4)
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title)
plt.colorbar()
tick_marks = np.arange(len(classes))
plt.xticks(tick_marks, classes, rotation=45)
plt.yticks(tick_marks, classes)
if normalize:
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
#print("Normalized confusion matrix")
else:
#print('Confusion matrix, without normalization')
pass
#print(cm)
label = [["\n True Negative", "\n False Positive \n Type II Error"],
["\n False Negative \n Type I Error", "\n True Positive"]
]
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, "{} {}".format(cm[i, j].round(4), label[i][j]),
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
def plot(actual_value, pred_value):
from sklearn.metrics import confusion_matrix
cm_2labels = confusion_matrix(y_pred = pred_value, y_true = actual_value)
plt.figure(figsize=[6,6])
plot_confusion_matrix(cm_2labels, ['Normal', 'Attack'], normalize = False)
In [15]:
past_scores = pd.read_pickle("dataset/scores/tf_lstm_nsl_kdd-orig_all.pkl")
In [16]:
past_scores.sort_values(by='f1_score',ascending=False)
Out[16]:
In [17]:
psg = past_scores.sort_values(by='test_score', ascending=False).groupby(by=['no_of_features', 'hidden_layers'])
psg.first().sort_values(by='test_score', ascending=False)
Out[17]:
In [18]:
psg.mean().sort_values(by='test_score', ascending=False)
Out[18]:
In [19]:
Train.predictions = pd.read_pickle("dataset/tf_lstm_nsl_kdd_predictions.pkl")
Train.predictions_ = pd.read_pickle("dataset/tf_lstm_nsl_kdd_predictions__.pkl")
In [20]:
#epoch_nof_hidden
Train.predictions["11_1_3"].sample()
Out[20]:
In [22]:
Train.predictions_["11_1_3"].sample()
Out[22]:
In [27]:
df = Train.predictions["11_1_3"].dropna()
df_ = Train.predictions_["11_1_3"].dropna()
In [28]:
from sklearn import metrics as me
def get_score(y_true, y_pred):
f1 = me.f1_score(y_true, y_pred)
pre = me.precision_score(y_true, y_pred)
rec = me.recall_score(y_true, y_pred)
acc = me.accuracy_score(y_true, y_pred)
return {"F1 Score":f1, "Precision":pre, "Recall":rec, "Accuracy":acc}
In [29]:
from sklearn import metrics as me
scores = get_score(df.loc[:,'Actual'].values.astype(int),
df.loc[:,'Prediction'].values.astype(int))
scores.update({"Scenario":"Train+/Test+"})
score_df = pd.DataFrame(scores, index=[0])
scores = get_score(df_.loc[:,'Actual'].values.astype(int),
df_.loc[:,'Prediction'].values.astype(int))
scores.update({"Scenario":"Train+/Test-"})
score_df = score_df.append(pd.DataFrame(scores, index=[1]))
score_df
Out[29]:
In [30]:
df.groupby(by="Actual").Actual.count()
Out[30]:
In [31]:
plot(actual_value = df.loc[:,'Actual'].values.astype(int),
pred_value = df.loc[:,'Prediction'].values.astype(int))
In [32]:
df_.groupby(by="Actual").Actual.count()
Out[32]:
In [33]:
plot(actual_value = df_.loc[:,'Actual'].values.astype(int),
pred_value = df_.loc[:,'Prediction'].values.astype(int))
In [34]:
from scipy import stats
def fn(x):
#print(x)
return stats.norm.interval(0.95, loc=x.f1_score.mean(), scale=x.f1_score.std())
psg.apply(fn)
Out[34]:
In [ ]:
In [ ]: