In [1]:
import tensorflow as tf
import pandas as pd
import numpy as np
pd.set_option("display.max_rows",15)
%matplotlib inline
sess = tf.InteractiveSession()
In [2]:
tf.reset_default_graph()
input_dim = 118
classes = 2
hidden_encoder_dim = 80
latent_dim = 10
hidden_decoder_dim = 80
lam = 0.01
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.001)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.01, shape=shape)
return tf.Variable(initial)
l2_loss = tf.constant(0.001)
#learning_rate = tf.Variable(initial_value=0.001)
with tf.variable_scope("Input"):
x = tf.placeholder("float", shape=[None, input_dim])
y_ = tf.placeholder("float", shape=[None, classes])
keep_prob = tf.placeholder("float")
with tf.variable_scope("Layer_Encoder"):
W_encoder_input_hidden = weight_variable([input_dim,hidden_encoder_dim])
b_encoder_input_hidden = bias_variable([hidden_encoder_dim])
l2_loss += tf.nn.l2_loss(W_encoder_input_hidden)
# Hidden layer encoder
hidden_encoder = tf.nn.relu(tf.matmul(x, W_encoder_input_hidden) + b_encoder_input_hidden)
tf.summary.histogram("Weights_Encoder", W_encoder_input_hidden)
hidden_encoder = tf.nn.dropout(hidden_encoder, keep_prob=keep_prob)
with tf.variable_scope("Layer_Mean"):
W_encoder_hidden_mu = weight_variable([hidden_encoder_dim,latent_dim])
b_encoder_hidden_mu = bias_variable([latent_dim])
l2_loss += tf.nn.l2_loss(W_encoder_hidden_mu)
# Mu encoder
mu_encoder = tf.matmul(hidden_encoder, W_encoder_hidden_mu) + b_encoder_hidden_mu
tf.summary.histogram("Weights_Mean", W_encoder_hidden_mu)
with tf.variable_scope("Layer_Variance"):
W_encoder_hidden_logvar = weight_variable([hidden_encoder_dim,latent_dim])
b_encoder_hidden_logvar = bias_variable([latent_dim])
l2_loss += tf.nn.l2_loss(W_encoder_hidden_logvar)
# Sigma encoder
logvar_encoder = tf.matmul(hidden_encoder, W_encoder_hidden_logvar) + b_encoder_hidden_logvar
tf.summary.histogram("Weights_Variance", W_encoder_hidden_logvar)
with tf.variable_scope("Sampling_Distribution"):
# Sample epsilon
epsilon = tf.random_normal(tf.shape(logvar_encoder), name='epsilon')
# Sample latent variable
std_encoder = tf.exp(0.5 * logvar_encoder)
z = mu_encoder + tf.multiply(std_encoder, epsilon)
tf.summary.histogram("Sample_Distribution", z)
with tf.variable_scope("Layer_Decoder"):
W_decoder_z_hidden = weight_variable([latent_dim,hidden_decoder_dim])
b_decoder_z_hidden = bias_variable([hidden_decoder_dim])
l2_loss += tf.nn.l2_loss(W_decoder_z_hidden)
# Hidden layer decoder
hidden_decoder = tf.nn.relu(tf.matmul(z, W_decoder_z_hidden) + b_decoder_z_hidden)
hidden_decoder = tf.nn.dropout(hidden_decoder, keep_prob=keep_prob)
tf.summary.histogram("Weights_Decoder", W_decoder_z_hidden)
with tf.variable_scope("Layer_Reconstruction"):
W_decoder_hidden_reconstruction = weight_variable([hidden_decoder_dim, input_dim])
b_decoder_hidden_reconstruction = bias_variable([input_dim])
l2_loss += tf.nn.l2_loss(W_decoder_hidden_reconstruction)
x_hat = tf.matmul(hidden_decoder, W_decoder_hidden_reconstruction) + b_decoder_hidden_reconstruction
tf.summary.histogram("Weights_Reconstruction", W_decoder_hidden_reconstruction)
with tf.variable_scope("Layer_Dense_Hidden"):
hidden_output = tf.layers.dense(z,latent_dim, activation=tf.nn.relu)
with tf.variable_scope("Layer_Dense_Softmax"):
y = tf.layers.dense(z, classes, activation=tf.nn.softmax)
with tf.variable_scope("Loss"):
BCE = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=x_hat, labels=x), reduction_indices=1)
KLD = -0.5 * tf.reduce_mean(1 + logvar_encoder - tf.pow(mu_encoder, 2) - tf.exp(logvar_encoder), reduction_indices=1)
softmax_loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels = y_, logits = y))
loss = tf.reduce_mean(BCE + KLD + softmax_loss)
regularized_loss = tf.abs(loss + lam * l2_loss, name = "Regularized_loss")
correct_prediction = tf.equal(tf.argmax(y_, 1), tf.argmax(y, 1))
tf_accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32), name = "Accuracy")
pred = tf.argmax(y, 1)
actual = tf.argmax(y_, 1)
#tf.summary.scalar("BCE", BCE)
#tf.summary.scalar("KLD", KLD)
#tf.summary.scalar("Softmax_loss", softmax_loss)
tf.summary.scalar("loss", regularized_loss)
with tf.variable_scope("Optimizer"):
learning_rate=0.001
grad_clip=5
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(regularized_loss, tvars), grad_clip)
train_op = tf.train.AdamOptimizer(learning_rate)
optimizer = train_op.apply_gradients(zip(grads, tvars))
# add op for merging summary
summary_op = tf.summary.merge_all()
# add Saver ops
# saver = tf.train.Saver()
In [3]:
kdd_train_2labels = pd.read_pickle("dataset/kdd_train_2labels_20percent.pkl")
kdd_test_2labels = pd.read_pickle("dataset/kdd_test_2labels_20percent.pkl")
output_columns_2labels = ['is_Attack','is_Normal']
from sklearn import model_selection as ms
from sklearn import preprocessing as pp
x_input = kdd_train_2labels.drop(output_columns_2labels, axis = 1)
y_output = kdd_train_2labels.loc[:,output_columns_2labels]
ss = pp.StandardScaler()
x_input = ss.fit_transform(x_input)
x_test = kdd_test_2labels.drop(output_columns_2labels, axis = 1)
y_test = kdd_test_2labels.loc[:,output_columns_2labels]
x_test = ss.transform(x_test)
In [4]:
epochs = 40
batch_iterations = 100
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())
for epoch in range(0, epochs):
x_train1, x_valid1, y_train1, y_valid1 = ms.train_test_split(x_input,
y_output.values,
test_size=0.2)
#x_valid, x_test, y_valid, y_test = ms.train_test_split(x_valid, y_valid, test_size = 0.4)
batch_indices = np.array_split(np.arange(x_train1.shape[0]), batch_iterations)
for i in batch_indices:
_, train_loss, summary_str = sess.run([optimizer, regularized_loss, summary_op],
feed_dict={x: x_train1[i,:], y_: y_train1[i,:], keep_prob:0.6})
summary_writer_train.add_summary(summary_str, epoch)
accuracy, summary_str = sess.run([tf_accuracy, summary_op],
feed_dict={x: x_valid1, y_:y_valid1, keep_prob:1})
summary_writer_valid.add_summary(summary_str, epoch)
if epoch % 2 == 0:
print("Step {} | Training Loss: {:.4f} | Validation Accuracy: {:.4f}".format(epoch, train_loss, accuracy))
accuracy, pred_value, actual_value = sess.run([tf_accuracy, pred, actual], feed_dict={x: x_test, y_:y_test, keep_prob:1})
print("Accuracy on Test data: {}".format(accuracy))
In [5]:
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')
print(cm)
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, cm[i, j].round(4),
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
In [6]:
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, output_columns_2labels, normalize = True)