Scribd
Upload
70 views10 pages

System Impl

The document loads libraries and imports for machine learning models in Keras and TensorFlow. It then loads and preprocesses URL data, splitting it into training and test sets. A 1D convolutional neural network model is defined to classify URLs as malicious or benign. The model is trained, evaluated on the test set, and the weights are saved. Functions for loading and saving the model are also defined.

Uploaded by

Sree Ram Pavan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
70 views10 pages

System Impl

The document loads libraries and imports for machine learning models in Keras and TensorFlow. It then loads and preprocesses URL data, splitting it into training and test sets. A 1D convolutional neural network model is defined to classify URLs as malicious or benign. The model is trained, evaluated on the test set, and the weights are saved. Functions for loading and saving the model are also defined.

Uploaded by

Sree Ram Pavan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 10

# Load Libraries - Make sure to run this cell!

import pandas as pd

import numpy as np

import re, os

from string import printable

from sklearn import model_selection

#import gensim

import tensorflow as tf

from keras.models import Sequential, Model, model_from_json, load_model

from keras import regularizers

from keras.layers.core import Dense, Dropout, Activation, Lambda, Flatten

from keras.layers import Input, ELU, LSTM, Embedding, Convolution2D, MaxPooling2D, \

BatchNormalization, Convolution1D, MaxPooling1D, concatenate

from keras.preprocessing import sequence

from keras.optimizers import SGD, Adam, RMSprop

from keras.utils import np_utils

from keras import backend as K

from pathlib import Path

import json

import warnings

warnings.filterwarnings("ignore")
2.

DATA_HOME = 'data/'

df = pd.read_csv(DATA_HOME + 'url_data_mega_deep_learning.csv')

df.sample(n=25).head(25)
out 2:

url_int_tokens = [[printable.index(x) + 1 for x in url if x in printable] for url in df.url]

max_len=75

X = sequence.pad_sequences(url_int_tokens, maxlen=max_len)

target = np.array(df.isMalicious)

print('Matrix dimensions of X: ', X.shape, 'Vector dimension of target: ', target.shape)

X_train, X_test, target_train, target_test = model_selection.train_test_split(X, target, test_size=0.25,


random_state=33)

def print_layers_dims(model):

l_layers = model.layers

# Note None is ALWAYS batch_size

for i in range(len(l_layers)):

print(l_layers[i])

print('Input Shape: ', l_layers[i].input_shape, 'Output Shape: ', l_layers[i].output_shape)

def save_model(fileModelJSON,fileWeights):

#print("Saving model to disk: ",fileModelJSON,"and",fileWeights)

#have h5py installed

if Path(fileModelJSON).is_file():

os.remove(fileModelJSON)

json_string = model.to_json()

with open(fileModelJSON,'w' ) as f:

json.dump(json_string, f)
if Path(fileWeights).is_file():

os.remove(fileWeights)

model.save_weights(fileWeights)

def load_model(fileModelJSON,fileWeights):

#print("Saving model to disk: ",fileModelJSON,"and",fileWeights)

with open(fileModelJSON, 'r') as f:

model_json = json.load(f)

model = model_from_json(model_json)

model.load_weights(fileWeights)

return model

def conv_fully(max_len=75, emb_dim=32, max_vocab_len=100, W_reg=regularizers.l2(1e-4)):

# Input

main_input = Input(shape=(max_len,), dtype='int32', name='main_input')

# Embedding layer

emb = Embedding(input_dim=max_vocab_len, output_dim=emb_dim, input_length=max_len,

W_regularizer=W_reg)(main_input)

emb = Dropout(0.25)(emb)

def sum_1d(X):

return K.sum(X, axis=1)

def get_conv_layer(emb, kernel_size=5, filters=256):

conv = Convolution1D(kernel_size=kernel_size, filters=filters, \

border_mode='same')(emb)
conv = ELU()(conv)

conv = Lambda(sum_1d, output_shape=(filters,))(conv)

#conv = BatchNormalization(mode=0)(conv)

conv = Dropout(0.5)(conv)

return conv

conv1 = get_conv_layer(emb, kernel_size=2, filters=256)

conv2 = get_conv_layer(emb, kernel_size=3, filters=256)

conv3 = get_conv_layer(emb, kernel_size=4, filters=256)

conv4 = get_conv_layer(emb, kernel_size=5, filters=256)

merged = concatenate([conv1,conv2,conv3,conv4], axis=1)

hidden1 = Dense(1024)(merged)

hidden1 = ELU()(hidden1)

hidden1 = BatchNormalization(mode=0)(hidden1)

hidden1 = Dropout(0.5)(hidden1)

hidden2 = Dense(1024)(hidden1)

hidden2 = ELU()(hidden2)

hidden2 = BatchNormalization(mode=0)(hidden2)

hidden2 = Dropout(0.5)(hidden2)

output = Dense(1, activation='sigmoid', name='output')(hidden2)

model = Model(input=[main_input], output=[output])

adam = Adam(lr=1e-4, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)

model.compile(optimizer=adam, loss='binary_crossentropy', metrics=['accuracy'])

return model
epochs = 5

batch_size = 32

model = conv_fully()

model.fit(X_train, target_train, epochs=epochs, batch_size=batch_size)

loss, accuracy = model.evaluate(X_test, target_test, verbose=1)

print('\nFinal Cross-Validation Accuracy', accuracy, '\n')

print_layers_dims(model)

target_proba = model.predict(X_test, batch_size=1)

target_proba[0:10]

model_name = "deeplearning_1DConv"

save_model(DATA_HOME + model_name + ".json", DATA_HOME + model_name + ".h5")

model = load_model(DATA_HOME + model_name + ".json", DATA_HOME + model_name + ".h5")

l_layers = model.layers

weights = l_layers[1].get_weights()

weights[0].shape
test_url_mal = "naureen.net/etisalat.ae/index2.php"

test_url_benign = "sixt.com/php/reservation?language=en_US"

url = test_url_benign

url_int_tokens = [[printable.index(x) + 1 for x in url if x in printable]]

max_len=75

X = sequence.pad_sequences(url_int_tokens, maxlen=max_len)

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy