Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import tensorflow as tf
|
| 3 |
+
import pickle
|
| 4 |
+
import numpy as np
|
| 5 |
+
import requests
|
| 6 |
+
from ProGPT import Conversation
|
| 7 |
+
|
| 8 |
+
# Load saved components
|
| 9 |
+
with open('preprocessing_params.pkl', 'rb') as f:
|
| 10 |
+
preprocessing_params = pickle.load(f)
|
| 11 |
+
with open('fisher_information.pkl', 'rb') as f:
|
| 12 |
+
fisher_information = pickle.load(f)
|
| 13 |
+
with open('label_encoder.pkl', 'rb') as f:
|
| 14 |
+
label_encoder = pickle.load(f)
|
| 15 |
+
with open('url_tokenizer.pkl', 'rb') as f:
|
| 16 |
+
url_tokenizer = pickle.load(f)
|
| 17 |
+
with open('html_tokenizer.pkl', 'rb') as f:
|
| 18 |
+
html_tokenizer = pickle.load(f)
|
| 19 |
+
|
| 20 |
+
# Load the model with custom loss
|
| 21 |
+
@tf.keras.utils.register_keras_serializable()
|
| 22 |
+
class EWCLoss(tf.keras.losses.Loss):
|
| 23 |
+
def __init__(self, model, fisher_information, importance=1.0, reduction='auto', name=None):
|
| 24 |
+
super(EWCLoss, self).__init__(reduction=reduction, name=name)
|
| 25 |
+
self.model = model
|
| 26 |
+
self.fisher_information = fisher_information
|
| 27 |
+
self.importance = importance
|
| 28 |
+
self.prev_weights = [layer.numpy() for layer in model.trainable_weights]
|
| 29 |
+
|
| 30 |
+
def call(self, y_true, y_pred):
|
| 31 |
+
standard_loss = tf.keras.losses.binary_crossentropy(y_true, y_pred)
|
| 32 |
+
ewc_loss = 0.0
|
| 33 |
+
for layer, fisher_info, prev_weight in zip(self.model.trainable_weights, self.fisher_information, self.prev_weights):
|
| 34 |
+
ewc_loss += tf.reduce_sum(fisher_info * tf.square(layer - prev_weight))
|
| 35 |
+
return standard_loss + (self.importance / 2.0) * ewc_loss
|
| 36 |
+
|
| 37 |
+
def get_config(self):
|
| 38 |
+
config = super().get_config()
|
| 39 |
+
config.update({
|
| 40 |
+
'importance': self.importance,
|
| 41 |
+
'reduction': self.reduction,
|
| 42 |
+
'name': self.name,
|
| 43 |
+
})
|
| 44 |
+
return config
|
| 45 |
+
|
| 46 |
+
@classmethod
|
| 47 |
+
def from_config(cls, config):
|
| 48 |
+
# Load fisher information from external file
|
| 49 |
+
with open('fisher_information.pkl', 'rb') as f:
|
| 50 |
+
fisher_information = pickle.load(f)
|
| 51 |
+
return cls(model=None, fisher_information=fisher_information, **config)
|
| 52 |
+
|
| 53 |
+
# Load the model
|
| 54 |
+
model = tf.keras.models.load_model('new_phishing_detection_model.keras',
|
| 55 |
+
custom_objects={'EWCLoss': EWCLoss})
|
| 56 |
+
|
| 57 |
+
# Recompile the model
|
| 58 |
+
ewc_loss = EWCLoss(model=model, fisher_information=fisher_information, importance=1000)
|
| 59 |
+
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.0005),
|
| 60 |
+
loss=ewc_loss,
|
| 61 |
+
metrics=['accuracy', tf.keras.metrics.Precision(), tf.keras.metrics.Recall()])
|
| 62 |
+
|
| 63 |
+
# Chatbot setup
|
| 64 |
+
access_token = 'your_pro_gpt_access_token'
|
| 65 |
+
chatbot = Conversation(access_token)
|
| 66 |
+
|
| 67 |
+
# Function to preprocess input
|
| 68 |
+
def preprocess_input(input_text, tokenizer, max_length):
|
| 69 |
+
sequences = tokenizer.texts_to_sequences([input_text])
|
| 70 |
+
padded_sequences = tf.keras.preprocessing.sequence.pad_sequences(sequences, maxlen=max_length, padding='post', truncating='post')
|
| 71 |
+
return padded_sequences
|
| 72 |
+
|
| 73 |
+
# Function to get prediction
|
| 74 |
+
def get_prediction(input_text, input_type):
|
| 75 |
+
is_url = input_type == "URL"
|
| 76 |
+
if is_url:
|
| 77 |
+
input_data = preprocess_input(input_text, url_tokenizer, preprocessing_params['max_url_length'])
|
| 78 |
+
else:
|
| 79 |
+
input_data = preprocess_input(input_text, html_tokenizer, preprocessing_params['max_html_length'])
|
| 80 |
+
|
| 81 |
+
prediction = model.predict([input_data, input_data])[0][0]
|
| 82 |
+
return prediction
|
| 83 |
+
|
| 84 |
+
# Function to fetch latest phishing sites from PhishTank
|
| 85 |
+
def fetch_latest_phishing_sites():
|
| 86 |
+
try:
|
| 87 |
+
response = requests.get('https://data.phishtank.com/data/online-valid.json')
|
| 88 |
+
data = response.json()
|
| 89 |
+
return data[:5]
|
| 90 |
+
except Exception as e:
|
| 91 |
+
return []
|
| 92 |
+
|
| 93 |
+
# Gradio UI
|
| 94 |
+
def phishing_detection(input_text, input_type):
|
| 95 |
+
prediction = get_prediction(input_text, input_type)
|
| 96 |
+
if prediction > 0.5:
|
| 97 |
+
return f"Warning: This site is likely a phishing site! ({prediction:.2f})"
|
| 98 |
+
else:
|
| 99 |
+
return f"Safe: This site is not likely a phishing site. ({prediction:.2f})"
|
| 100 |
+
|
| 101 |
+
def latest_phishing_sites():
|
| 102 |
+
sites = fetch_latest_phishing_sites()
|
| 103 |
+
return [f"{site['url']}" for site in sites]
|
| 104 |
+
|
| 105 |
+
def chatbot_response(user_input):
|
| 106 |
+
response = chatbot.prompt(user_input)
|
| 107 |
+
return response
|
| 108 |
+
|
| 109 |
+
iface = gr.Interface(
|
| 110 |
+
fn=phishing_detection,
|
| 111 |
+
inputs=[gr.inputs.Textbox(lines=5, placeholder="Enter URL or HTML code"), gr.inputs.Radio(["URL", "HTML"], type="value", label="Input Type")],
|
| 112 |
+
outputs="text",
|
| 113 |
+
title="Phishing Detection with Enhanced EWC Model",
|
| 114 |
+
description="Check if a URL or HTML is Phishing",
|
| 115 |
+
theme="default"
|
| 116 |
+
)
|
| 117 |
+
|
| 118 |
+
iface.launch()
|