import pandas as pd
import numpy as np
import glob
from sklearn.model_selection import train_test_split
import SeperateData
import Vectorizer
import TFIDFVectorizer
import PreprocessData
import MLAlgorithms
import warnings
import AdversarialAttack
#import AdversarialMLAttack

# Ignore all warnings
warnings.filterwarnings("ignore")

# Specify the folder containing the CSV files (modify this path accordingly)
folder_path = "URL Datasets/MLClassification/*.csv"  # Example: "data/*.csv"

class MaliciousURLS:
    # Get a list of all CSV files in the folder
    csv_files = glob.glob(folder_path)

    # Read and combine all CSV files into a single DataFrame
    url_df = pd.concat((pd.read_csv(file, index_col=0) for file in csv_files), ignore_index=True)
    
    #seperate the training and test data
    test_percentage = .2
    train_df, test_df = train_test_split(url_df, test_size=test_percentage, random_state=42)
    seperated_Traindf,seperated_Testdf=SeperateData.dataSeperation(train_df, test_df)
    #print(seperated_Traindf[0].head)
    
    # Convert y_train and y_test to 1D array
    y_train = seperated_Traindf[6].values.ravel() if isinstance(seperated_Traindf[6], pd.DataFrame) else seperated_Traindf[6]
    y_test = seperated_Testdf[6].values.ravel() if isinstance(seperated_Testdf[6], pd.DataFrame) else seperated_Testdf[6]
    
    #Block1: NLP Based Approach
    # Domain Vectorization
    
    traincountvector_X, testcountvector_X =Vectorizer.countVectorizer(seperated_Traindf[0], seperated_Testdf[0])
    print("countVectorizing Completed")
    # traintfidfvector_X, testtfidfvector_X  =Vectorizer.TfidVectorizer(seperated_Traindf[0], seperated_Testdf[0])
    # print("TFIDFVectorizing Completed")
    # trainword2vector_X, testword2vector_X  =Vectorizer.TfidVectorizer(seperated_Traindf[0], seperated_Testdf[0])
    # print("Word2VecVectorizing Completed")
    # Evaluate adversarial attacks on all trained NLP models

    #Train the  NLP Model
    #method_name={0:"Bag of Words",1: "TF-IDF",2: "Word2Vec"}
    #model_accuracies, trained_models = MLAlgorithms.train_NLP(traincountvector_X, testcountvector_X, y_train, y_test,method_name[1])
    #model_accuracies = MLAlgorithms.train_NLP(traintfidfvector_X, testtfidfvector_X, y_train, y_test,method_name[1])
    #model_accuracies = MLAlgorithms.train_NLP(trainword2vector_X, testword2vector_X, y_train, y_test,method_name[2])
    #print(trained_models)
    #attack_results = AdversarialAttack.apply_adversarial_attackUp(trained_models['Naive Bayes (Multinomial)'], testcountvector_X, y_test, method_name[1])
    
    #Block2: Convert URL into different feature
    method_name={1:'LexicalFeature', 2:'LinguisticFeature', 3: 'Feature', 4: 'HostBasedFeature', 5: 'ContentBasedFeature'}
    trainlexicialvector_X =PreprocessData.DataPreparationUp(seperated_Traindf[3])
    testlexicialvector_X =PreprocessData.DataPreparationUp(seperated_Testdf[3])
    # print(seperated_Traindf[3])
    # print(trainlexicialvector_X)
    #HostBasedFeature.process_domains(input_filename, output_filename)
    #print("HostBasedFeature Completed")
    # traindictonaryvector_X, testdictonaryvector_X =DictonaryWord.dictionary_word(train_df, test_df)
    # print("Linguistic(DictonaryWord) Completed")
    # traindictonaryvector_X, testdictonaryvector_X =ContentBasedFeature.ContentBasedFeature(train_df, test_df)
    # print("Content-based Feature Vectorizing Completed")
    
    # model train
    #MLAlgorithms.LogisticReg(trainlexicialvector_X, y_train, testlexicialvector_X, y_test)
    model_accuracies,trained_models = MLAlgorithms.train_classifiers(trainlexicialvector_X, testlexicialvector_X, y_train, y_test, method_name[3])
    
    # Apply adversarial attacks and compare model robustness
    #attack_results = AdversarialMLAttack.evaluate_adversarial_attacks(trained_models, testlexicialvector_X, y_test)